Upload 89 files

.gitattributes

@@ -33,3 +33,17 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+kaggle_output/drift_simulation.png filter=lfs diff=lfs merge=lfs -text
+kaggle_output/eda_overview.png filter=lfs diff=lfs merge=lfs -text
+kaggle_output/models/lgbm_fold_3.txt filter=lfs diff=lfs merge=lfs -text
+kaggle_output/models/lgbm_fold_4.txt filter=lfs diff=lfs merge=lfs -text
+kaggle_output/models/lgbm_fold_5.txt filter=lfs diff=lfs merge=lfs -text
+kaggle_output/river_drift_detection.png filter=lfs diff=lfs merge=lfs -text
+kaggle_output/shap_bar.png filter=lfs diff=lfs merge=lfs -text
+kaggle_output/shap_beeswarm.png filter=lfs diff=lfs merge=lfs -text
+kaggle_output/shap_dependence.png filter=lfs diff=lfs merge=lfs -text
+kaggle_output/shap_waterfall.png filter=lfs diff=lfs merge=lfs -text
+models/lgbm_fold_3.txt filter=lfs diff=lfs merge=lfs -text
+models/lgbm_fold_4.txt filter=lfs diff=lfs merge=lfs -text
+models/lgbm_fold_5.txt filter=lfs diff=lfs merge=lfs -text
+utils/2.jpeg filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,207 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+#poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+#pdm.lock
+#pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+#pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Abstra
+# Abstra is an AI-powered process automation framework.
+# Ignore directories containing user credentials, local state, and settings.
+# Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#  Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#  that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#  and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#  you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Cursor
+#  Cursor is an AI-powered code editor. `.cursorignore` specifies files/directories to
+#  exclude from AI features like autocomplete and code analysis. Recommended for sensitive data
+#  refer to https://docs.cursor.com/context/ignore-files
+.cursorignore
+.cursorindexingignore
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/

README.md

@@ -1,20 +1,357 @@
+<p align="center">
+  <h1 align="center">💳 Credit Invisibility Solver</h1>
+  <p align="center">
+    <b>Explainable Credit Risk Modeling with Alternative Data, NLP Embeddings & Concept Drift Detection</b>
+  </p>
+  <p align="center">
+    <img src="https://img.shields.io/badge/Model-LightGBM%20%2B%20XGBoost-brightgreen" />
+    <img src="https://img.shields.io/badge/NLP-Sentence--BERT-blue" />
+    <img src="https://img.shields.io/badge/Explainability-SHAP-orange" />
+    <img src="https://img.shields.io/badge/Drift-River%20ADWIN-red" />
+    <img src="https://img.shields.io/badge/Tracking-W%26B-yellow" />
+    <img src="https://img.shields.io/badge/Deploy-Streamlit-ff4b4b" />
+  </p>
+</p>
+
+---
+
+## Problem Statement
+
+**1.7 billion adults worldwide are credit-invisible** — they have no formal credit history, locking them out of loans, insurance, and financial services. Traditional credit scoring relies on bureau data that simply doesn't exist for these populations.
+
+This project builds an **end-to-end ML pipeline** that:
+1. **Engineers 200+ features** from 7 relational tables (Home Credit Default Risk dataset)
+2. **Fuses tabular + NLP signals** using Sentence-BERT embeddings of synthesized financial narratives
+3. **Trains an optimized LightGBM + XGBoost ensemble** with Optuna hyperparameter tuning
+4. **Explains every prediction** with SHAP (beeswarm, waterfall, dependence plots)
+5. **Detects concept drift** in real-time using River's ADWIN detector
+6. **Deploys as an interactive Streamlit dashboard** for instant credit scoring
+
+---
+
+## 🏗️ Architecture
+
+```
+┌─────────────────────────────────────────────────────────────────────┐
+│                    7 Raw Tables (Home Credit)                       │
+│  application_train/test │ bureau │ prev_app │ installments │ ...    │
+└──────────────┬──────────────────────────────────────────────────────┘
+               │
+    ┌──────────▼──────────┐     ┌──────────────────────┐
+    │  Feature Engineering │     │  NLP Pipeline         │
+    │  • Bureau aggregates │     │  • Financial narrative │
+    │  • Prev app signals  │     │    synthesis           │
+    │  • Installment       │     │  • Sentence-BERT       │
+    │    behaviour          │     │    encoding            │
+    │  • POS Cash / CC     │     │  • PCA → 32 dims       │
+    │  • Domain ratios     │     └──────────┬─────────────┘
+    └──────────┬──────────┘                │
+               │          ┌────────────────┘
+               ▼          ▼
+    ┌──────────────────────────┐
+    │  Merged Feature Matrix    │
+    │  207 features total       │
+    │  (175 tabular + 32 NLP)   │
+    └──────────┬───────────────┘
+               │
+    ┌──────────▼──────────┐
+    │  Optuna HPO           │
+    │  80 trials each       │
+    │  TPE + MedianPruner   │
+    └──────────┬───────────┘
+               │
+    ┌──────────▼──────────────────────┐
+    │  Ensemble: 0.6×LightGBM + 0.4×XGBoost │
+    │  5-Fold Stratified CV                    │
+    │  + Logistic Regression Blending          │
+    └──────────┬──────────────────────────────┘
+               │
+    ┌──────────▼──────────┐     ┌──────────────────────┐
+    │  SHAP Explainability │     │  River ADWIN Drift    │
+    │  • TreeExplainer     │     │  • Online learning     │
+    │  • Beeswarm / Bar    │     │  • Auto-retrain        │
+    │  • Waterfall         │     │  • Drift simulation    │
+    └──────────────────────┘     └──────────────────────┘
+               │
+    ┌──────────▼──────────┐
+    │  Streamlit Dashboard  │
+    │  • Live scoring       │
+    │  • SHAP per applicant │
+    │  • Drift sensitivity  │
+    └───────────────────────┘
+```
+
+---
+
+##  Streamlit Dashboard
+
+### Score Breakdown — Gauge Chart + Risk Factor Radar
+
+![Score Breakdown](utils/1.jpeg)
+
+### SHAP Explainability — Top 15 Feature Contributions
+
+![SHAP Explainability](utils/2.jpeg)
+
+### Drift Simulation — Income Shock Sensitivity
+
+![Drift Simulation](utils/3.jpeg)
+
+### Feature Profile — Applicant Summary Table
+
+![Feature Profile](utils/4.jpeg)
+
 ---
-title: Explainable Credit Risk Modeling With Schduling
-emoji: 🚀
-colorFrom: red
-colorTo: red
-sdk: docker
-app_port: 8501
-tags:
-- streamlit
-pinned: false
-short_description: Streamlit template space
-license: mit
+
+## Project Structure
+
+```
+├── notebook.ipynb                  ← Kaggle training notebook (18 cells)
+├── streamlit_app.py                ← Interactive deployment dashboard
+├── src/
+│   ├── feature_engineering.py      ← 7-table feature pipeline
+│   ├── drift_detector.py           ← ADWIN/KSWIN drift detection + River online learner
+│   └── nlp_features.py             ← Sentence-BERT embedding pipeline
+├── models/                         ← Saved model artifacts
+│   ├── lgbm_fold_{1-5}.txt         ← 5 LightGBM fold models
+│   ├── xgb_fold_{1-5}.json         ← 5 XGBoost fold models
+│   ├── pca.pkl                     ← Fitted PCA for NLP embeddings
+│   ├── scaler.pkl                  ← Fitted StandardScaler
+│   └── feature_cols.json           ← 207 feature column names
+├── kaggle_output/                  ← Full Kaggle run artifacts (plots, submission, logs)
+├── requirements.txt
+└── README.md
+```
+
 ---
 
-# Welcome to Streamlit!
+## EDA & Training Results
+
+### Exploratory Data Analysis
+
+![EDA Overview](kaggle_output/eda_overview.png)
+
+**Key findings:**
+- **91.9% non-default vs 8.1% default** — severe class imbalance (11.4:1 ratio)
+- Income distributions overlap heavily between defaulters and non-defaulters
+- Age is weakly predictive — younger applicants default slightly more
+- Occupation type shows strong signal (Laborers, Drivers have highest default rates)
+- 40%+ missing values in housing and employment-related columns
+
+---
+
+### SHAP Feature Importance
+
+#### Global Feature Importance (Mean |SHAP|)
+
+![SHAP Bar Plot](kaggle_output/shap_bar.png)
+
+**Top predictive features:** `EXT_SOURCE_MEAN`, `EXT_SOURCE_2`, `DAYS_BIRTH` (age), `CREDIT_INCOME_RATIO`, and `DAYS_EMPLOYED` dominate the model's decisions.
+
+#### SHAP Beeswarm — Per-Feature Impact Distribution
+
+![SHAP Beeswarm](kaggle_output/shap_beeswarm.png)
+
+Each dot represents one applicant. Red = high feature value, Blue = low. Features like `EXT_SOURCE_MEAN` show a clear trend: **higher external scores → lower default risk**.
+
+#### SHAP Dependence Plots — Top 3 Features
+
+![SHAP Dependence](kaggle_output/shap_dependence.png)
+
+Non-linear relationships revealed by SHAP dependence: external scores have diminishing returns beyond 0.7, and credit-to-income ratio inflects sharply above 3x.
+
+#### SHAP Waterfall — Highest-Risk Applicant
+
+![SHAP Waterfall](kaggle_output/shap_waterfall.png)
+
+Per-applicant explanation showing exactly how each feature pushed the prediction above/below the base rate. This is the core of the "explainability" promise.
+
+---
+
+### Concept Drift Simulation
+
+#### Batch Drift Scenarios
+
+![Drift Simulation](kaggle_output/drift_simulation.png)
+
+Simulated economic shocks (income reduction, mass job loss) show model AUC degradation. Under a **60% income shock**, AUC drops significantly, validating the need for drift detection.
+
+#### River ADWIN Online Drift Detection
+
+![River Drift Detection](kaggle_output/river_drift_detection.png)
+
+The ADWIN detector correctly identifies the injected concept drift at sample ~185k. The adaptive Hoeffding Tree auto-retrains on detection, showing the cumulative drift event count.
+
+---
+
+## Feature Engineering Pipeline
+
+### 7 Source Tables → 207 Features
 
-Edit `/src/streamlit_app.py` to customize this app to your heart's desire. :heart:
+| Source Table | Features | Key Signals |
+|---|---|---|
+| **Application** | Domain ratios, external scores, age/employment, document flags | `CREDIT_INCOME_RATIO`, `EXT_SOURCE_MEAN`, `AGE_YEARS` |
+| **Bureau** | Credit history aggregates, DPD rates, utilization | `BUREAU_DEBT_CREDIT_RATIO_MAX`, `BUREAU_ACTIVE_COUNT` |
+| **Previous Apps** | Approval/refusal rates, application patterns | `PREV_APPROVED_RATE`, `PREV_APP_CREDIT_RATIO_MEAN` |
+| **Installments** | Payment behaviour, late/short payment rates | `INST_LATE_PAYMENT_RATE`, `INST_PAYMENT_DIFF_MEAN` |
+| **POS Cash** | Point-of-sale DPD patterns | `POS_DPD_RATE`, `POS_SK_DPD_MAX` |
+| **Credit Card** | Utilization rates, drawing behaviour | `CC_UTIL_RATE_MEAN`, `CC_DRAWING_RATE_MEAN` |
+| **NLP Embeddings** | Sentence-BERT + PCA (32 dims) | `NLP_EMB_0` through `NLP_EMB_31` |
+
+### NLP Feature Pipeline
+
+Financial narratives are **synthesized from tabular signals** (in production, these would come from real user survey text, app usage data, or financial literacy assessments):
+
+```
+"Applicant aged 35 years with annual income of 250000 currency units.
+ Requesting credit of 500000 for personal needs. Employed for 5.0 years.
+ Client demonstrates moderate financial awareness with occasional late payments.
+ External credit assessment score: 0.55. Owns property which serves as collateral."
+```
+
+These are encoded with **Sentence-BERT (all-MiniLM-L6-v2)** and reduced to 32 dimensions via PCA, capturing semantic credit signals.
+
+---
+
+## Model Training
+
+### Ensemble Strategy
+
+| Component | Method | OOF AUC |
+|---|---|---|
+| **LightGBM** | 5-fold CV, Optuna-tuned (80 trials) | ~0.78 |
+| **XGBoost** | 5-fold CV, Optuna-tuned (80 trials) | ~0.77 |
+| **Ensemble** | 0.6×LGBM + 0.4×XGB weighted blend | ~0.79 |
+
+### Blending with Logistic Regression
+
+The ensemble uses a **fixed 60/40 weighted average** of LightGBM and XGBoost OOF predictions. In an extended pipeline, a **Logistic Regression meta-learner** can be stacked on top of the base model predictions:
+
+```python
+from sklearn.linear_model import LogisticRegression
+
+# Stack OOF predictions as meta-features
+meta_X = np.column_stack([oof_lgbm, oof_xgb])
+meta_lr = LogisticRegression(C=1.0)
+meta_lr.fit(meta_X, y)
+
+# Final blend = LR(lgbm_pred, xgb_pred)
+test_blend = meta_lr.predict_proba(np.column_stack([test_lgbm, test_xgb]))[:, 1]
+```
+
+This learns the optimal blending weights from data rather than using fixed 60/40.
+
+### Optuna Hyperparameter Optimization
+
+- **Sampler:** TPE (Tree-Structured Parzen Estimator)
+- **Pruner:** MedianPruner with 10 warmup steps — kills bad trials early
+- **Search space:** `num_leaves`, `learning_rate`, `feature_fraction`, `bagging_fraction`, `reg_alpha/lambda`, `max_depth`, `min_gain_to_split`
+
+---
+
+## Concept Drift Detection
+
+### Why Drift Matters
+
+Credit models degrade over time as economic conditions change. A model trained on pre-pandemic data won't perform well during a recession. This project implements:
+
+1. **Batch drift simulation** — apply synthetic income shocks (30-70% reduction) and measure AUC degradation
+2. **Online drift detection** — River's ADWIN detector monitors the prediction error stream in real-time
+3. **Auto-retrain** — when ADWIN fires, the Hoeffding Adaptive Tree resets with a faster learning rate
+
+### River Pipeline
+
+```python
+# Online pipeline: StandardScaler → Hoeffding Adaptive Tree
+pipeline = StandardScaler() | HoeffdingAdaptiveTreeClassifier(grace_period=200)
+
+# ADWIN monitors error stream
+adwin = ADWIN(delta=0.002)  # lower delta = more sensitive
+
+# On drift detection → rebuild pipeline with faster adaptation
+if adwin.drift_detected:
+    pipeline = StandardScaler() | HoeffdingAdaptiveTreeClassifier(grace_period=50)
+```
+
+---
+
+## Quick Start
+
+### Prerequisites
+
+- Python 3.10+
+- [uv](https://docs.astral.sh/uv/) (recommended) or pip
+
+### Setup
+
+```bash
+# Clone
+git clone https://github.com/suvraadeep/Explainable-Credit-Risk-Modeling-with-Schduling.git
+cd Explainable-Credit-Risk-Modeling-with-Schduling
+
+# Create venv and install deps
+uv venv .venv
+uv pip install --python .venv/Scripts/python.exe -r requirements.txt
+
+# Or with pip
+python -m venv .venv
+.venv/Scripts/activate    # Windows
+pip install -r requirements.txt
+```
+
+### Run the Streamlit Dashboard
+
+```bash
+# Windows
+.venv\Scripts\streamlit.exe run app.py
+
+# Linux/Mac
+.venv/bin/streamlit run app.py
+```
+
+Open **http://localhost:8501** in your browser.
+
+### Kaggle Notebook
+
+The full training pipeline runs on Kaggle with the [Home Credit Default Risk](https://www.kaggle.com/c/home-credit-default-risk) dataset. Upload the notebook and run all 18 cells to reproduce:
+- Feature engineering across 7 tables
+- Sentence-BERT NLP embeddings
+- Optuna HPO for LightGBM and XGBoost
+- 5-fold ensemble training
+- SHAP explainability suite
+- River online drift detection
+- W&B experiment logging
+
+---
+
+## W&B Experiment Tracking
+
+All experiments are tracked with [Weights & Biases](https://wandb.ai/):
+
+| Run | Metrics Logged |
+|---|---|
+| `lgbm-baseline` | Per-fold AUC, feature importance table |
+| `ensemble-lgbm-xgb` | Per-fold LightGBM/XGBoost/Ensemble AUC |
+| `concept-drift-simulation` | AUC under 5 economic shock scenarios |
+| `final-summary` | Consolidated metrics, artifact upload |
+
+Set your API key:
+```bash
+# Kaggle → Secrets → WANDB_API_KEY
+# Or in notebook:
+import wandb
+wandb.login()
+```
+
+---
+
+## License
+
+This project is licensed under the MIT License — see the [LICENSE](LICENSE) file for details.
+
+---
 
-If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
-forums](https://discuss.streamlit.io).
+<p align="center">
+  <b>Built for the 1.7B credit-invisible 🌍</b><br>
+  <sub>LightGBM + XGBoost + Sentence-BERT + SHAP + River (ADWIN) + W&B</sub>
+</p>

app.py

@@ -0,0 +1,431 @@
+# streamlit_app.py
+"""
+Credit Invisibility Solver — Streamlit App
+Run: streamlit run streamlit_app.py
+"""
+
+import streamlit as st
+import numpy as np
+import pandas as pd
+import shap
+import lightgbm as lgb
+import matplotlib.pyplot as plt
+import matplotlib
+matplotlib.use("Agg")
+import json
+import joblib
+import os
+import plotly.graph_objects as go
+from sentence_transformers import SentenceTransformer
+from sklearn.decomposition import PCA
+
+# ─── Page config ──────────────────────────────────────────────────────────────
+st.set_page_config(
+    page_title="Credit Invisibility Solver",
+    page_icon="💳",
+    layout="wide",
+    initial_sidebar_state="expanded",
+)
+
+# ─── Custom CSS ───────────────────────────────────────────────────────────────
+st.markdown("""
+<style>
+    .main-header { font-size: 2.4rem; font-weight: 800; color: #1565C0; text-align: center; padding: 1rem 0; }
+    .metric-card { background: linear-gradient(135deg, #1565C0, #42A5F5); border-radius: 12px;
+                   padding: 1.2rem; color: white; text-align: center; }
+    .risk-high   { background: #FFEBEE; border-left: 5px solid #F44336; padding: 1rem; border-radius: 8px; }
+    .risk-medium { background: #FFF8E1; border-left: 5px solid #FF9800; padding: 1rem; border-radius: 8px; }
+    .risk-low    { background: #E8F5E9; border-left: 5px solid #4CAF50; padding: 1rem; border-radius: 8px; }
+    .sidebar-section { font-size: 0.9rem; color: #666; margin-bottom: 0.4rem; font-weight: 600; }
+</style>
+""", unsafe_allow_html=True)
+
+# ─── Load artifacts ───────────────────────────────────────────────────────────
+MODEL_DIR = "./models"
+
+@st.cache_resource
+def load_models():
+    models = []
+    for i in range(1, 6):
+        path = f"{MODEL_DIR}/lgbm_fold_{i}.txt"
+        if os.path.exists(path):
+            m = lgb.Booster(model_file=path)
+            models.append(m)
+    return models
+
+@st.cache_resource
+def load_artifacts():
+    pca    = joblib.load(f"{MODEL_DIR}/pca.pkl")     if os.path.exists(f"{MODEL_DIR}/pca.pkl")    else None
+    scaler = joblib.load(f"{MODEL_DIR}/scaler.pkl")  if os.path.exists(f"{MODEL_DIR}/scaler.pkl") else None
+    fc_path = f"{MODEL_DIR}/feature_cols.json"
+    if os.path.exists(fc_path):
+        with open(fc_path) as f:
+            feature_cols = json.load(f)
+    else:
+        feature_cols = []
+    return pca, scaler, feature_cols
+
+@st.cache_resource
+def load_sbert():
+    try:
+        return SentenceTransformer("all-MiniLM-L6-v2")
+    except Exception:
+        return None
+
+# ─── Helper functions ─────────────────────────────────────────────────────────
+def build_single_applicant_features(inputs: dict, pca, sbert) -> pd.DataFrame:
+    """Transform raw user inputs into model-ready features."""
+    income  = inputs["income"]
+    credit  = inputs["credit_amount"]
+    age     = inputs["age"]
+    emp_yrs = inputs["employment_years"]
+    ext1    = inputs["ext_score_1"]
+    ext2    = inputs["ext_score_2"]
+    ext3    = inputs["ext_score_3"]
+    
+    # Build a synthetic text for NLP embedding
+    literacy = "strong financial planning habits" if np.mean([ext1,ext2,ext3]) > 0.6 else (
+               "moderate financial awareness" if np.mean([ext1,ext2,ext3]) > 0.4 else
+               "limited financial experience")
+    text = (
+        f"Applicant aged {age:.0f} years with annual income of {income:.0f}. "
+        f"Requesting credit of {credit:.0f}. Employed for {emp_yrs:.1f} years. "
+        f"Client demonstrates {literacy}. External score: {np.mean([ext1,ext2,ext3]):.2f}. "
+        f"{'Owns property.' if inputs['owns_realty'] else 'No property.'} "
+        f"{'Has dependents.' if inputs['has_children'] else 'No children.'}"
+    )
+    
+    # NLP embed + PCA
+    if sbert is not None:
+        emb = sbert.encode([text], normalize_embeddings=True)
+        if pca is not None:
+            emb = pca.transform(emb)
+        nlp_dict = {f"NLP_EMB_{i}": emb[0][i] for i in range(emb.shape[1])}
+    else:
+        # Demo mode — deterministic pseudo-embeddings
+        n_dims = pca.n_components_ if pca is not None else 32
+        nlp_dict = {f"NLP_EMB_{i}": 0.0 for i in range(n_dims)}
+    
+    # Tabular features
+    tab_dict = {
+        "AMT_INCOME_TOTAL":       income,
+        "AMT_CREDIT":             credit,
+        "AMT_ANNUITY":            inputs["annuity"],
+        "AMT_GOODS_PRICE":        credit * 0.9,
+        "DAYS_BIRTH":             -age * 365,
+        "DAYS_EMPLOYED":          -emp_yrs * 365,
+        "EXT_SOURCE_1":           ext1,
+        "EXT_SOURCE_2":           ext2,
+        "EXT_SOURCE_3":           ext3,
+        "EXT_SOURCE_MEAN":        np.mean([ext1, ext2, ext3]),
+        "EXT_SOURCE_MIN":         np.min([ext1, ext2, ext3]),
+        "EXT_SOURCE_PROD":        ext1 * ext2 * ext3,
+        "EXT_SOURCE_STD":         np.std([ext1, ext2, ext3]),
+        "EXT1_EXT2_INTERACTION":  ext1 * ext2,
+        "EXT2_EXT3_INTERACTION":  ext2 * ext3,
+        "CREDIT_INCOME_RATIO":    credit / (income + 1),
+        "ANNUITY_INCOME_RATIO":   inputs["annuity"] / (income + 1),
+        "CREDIT_TERM":            inputs["annuity"] / (credit + 1),
+        "AGE_YEARS":              age,
+        "EMPLOYMENT_YEARS":       emp_yrs,
+        "EMPLOYED_RATIO":         emp_yrs / (age + 1),
+        "INCOME_PER_PERSON":      income / (inputs["family_size"] + 1),
+        "CNT_FAM_MEMBERS":        inputs["family_size"],
+        "CNT_CHILDREN":           inputs["n_children"],
+        "CHILDREN_RATIO":         inputs["n_children"] / (inputs["family_size"] + 1),
+        "FLAG_OWN_REALTY":        int(inputs["owns_realty"]),
+        "FLAG_OWN_CAR":           int(inputs["owns_car"]),
+        "HAS_CAR_REALTY":         int(inputs["owns_realty"] and inputs["owns_car"]),
+        "DOCUMENT_COUNT":         inputs["doc_count"],
+        "TOTAL_ENQUIRIES":        inputs["total_enquiries"],
+        "BUREAU_COUNT":           inputs["bureau_count"],
+        "BUREAU_ACTIVE_COUNT":    inputs["bureau_active"],
+    }
+    
+    feat = {**tab_dict, **nlp_dict}
+    return pd.DataFrame([feat])
+
+def predict_risk(df_feat: pd.DataFrame, models: list, feature_cols: list) -> float:
+    """Ensemble predict across all loaded fold models."""
+    # Align columns — fill missing with 0
+    for col in feature_cols:
+        if col not in df_feat.columns:
+            df_feat[col] = 0.0
+    df_feat = df_feat[feature_cols]
+    preds = [m.predict(df_feat, num_iteration=m.best_iteration) for m in models]
+    return float(np.mean(preds))
+
+def risk_band(score: float) -> tuple:
+    if score < 0.15:
+        return "LOW RISK", "risk-low", "#4CAF50", "✅"
+    elif score < 0.40:
+        return "MEDIUM RISK", "risk-medium", "#FF9800", "⚠️"
+    else:
+        return "HIGH RISK", "risk-high", "#F44336", "🚨"
+
+def get_shap_values(model, df_feat, feature_cols):
+    for col in feature_cols:
+        if col not in df_feat.columns:
+            df_feat[col] = 0.0
+    df_feat = df_feat[feature_cols]
+    explainer = shap.TreeExplainer(model)
+    sv = explainer.shap_values(df_feat)
+    if isinstance(sv, list):
+        sv = sv[1]
+    return sv, explainer.expected_value if not isinstance(explainer.expected_value, list) else explainer.expected_value[1], df_feat
+
+# ─── Main App ─────────────────────────────────────────────────────────────────
+def main():
+    st.markdown('<h1 class="main-header">💳 Credit Invisibility Solver</h1>', unsafe_allow_html=True)
+    st.markdown(
+        "<p style='text-align:center; color:#555; font-size:1.1rem;'>"
+        "Alternative data ML pipeline to score the 1.7B credit-invisible population"
+        "</p>", unsafe_allow_html=True
+    )
+    st.divider()
+
+    # Load models
+    try:
+        models       = load_models()
+        pca, scaler, feature_cols = load_artifacts()
+        sbert        = load_sbert()
+        model_loaded = len(models) > 0
+    except Exception as e:
+        st.error(f"⚠️ Could not load models: {e}. Running in demo mode.")
+        model_loaded = False
+        models, pca, scaler, feature_cols = [], None, None, []
+
+    # ── Sidebar ───────────────────────────────────────────────────────────────
+    with st.sidebar:
+        st.image("https://img.shields.io/badge/Model-LightGBM%20%2B%20XGBoost-brightgreen", use_container_width=True)
+        st.markdown("### 🎛️ Applicant Profile")
+        
+        st.markdown('<div class="sidebar-section">Financial Info</div>', unsafe_allow_html=True)
+        income         = st.number_input("Annual Income (₹)", 10000, 10000000, 250000, step=10000)
+        credit_amount  = st.number_input("Requested Credit (₹)", 10000, 5000000, 500000, step=10000)
+        annuity        = st.number_input("Monthly Annuity (₹)", 1000, 200000, 15000, step=1000)
+        
+        st.markdown('<div class="sidebar-section">Personal Info</div>', unsafe_allow_html=True)
+        age            = st.slider("Age (years)", 20, 70, 35)
+        employment_yrs = st.slider("Employment Years", 0, 40, 5)
+        family_size    = st.slider("Family Size", 1, 10, 3)
+        n_children     = st.slider("Number of Children", 0, 5, 0)
+        
+        st.markdown('<div class="sidebar-section">Assets</div>', unsafe_allow_html=True)
+        owns_realty    = st.checkbox("Owns Property", True)
+        owns_car       = st.checkbox("Owns Car", False)
+        
+        st.markdown('<div class="sidebar-section">Credit Bureau Signals</div>', unsafe_allow_html=True)
+        ext_score_1    = st.slider("External Score 1 (Bureau)", 0.0, 1.0, 0.6, 0.01)
+        ext_score_2    = st.slider("External Score 2 (Behaviour)", 0.0, 1.0, 0.55, 0.01)
+        ext_score_3    = st.slider("External Score 3 (Alt Data)", 0.0, 1.0, 0.50, 0.01)
+        bureau_count   = st.number_input("# Previous Bureau Enquiries", 0, 50, 2)
+        bureau_active  = st.number_input("# Active Bureau Credits", 0, 20, 1)
+        total_enquiries= st.number_input("# Total Loan Enquiries", 0, 100, 3)
+        doc_count      = st.number_input("# Documents Submitted", 0, 20, 5)
+        
+        predict_btn    = st.button("🔮 Score Applicant", use_container_width=True, type="primary")
+
+    # ── Main Panels ───────────────────────────────────────────────────────────
+    col1, col2, col3 = st.columns(3)
+
+    inputs = dict(
+        income=income, credit_amount=credit_amount, annuity=annuity,
+        age=age, employment_years=employment_yrs, family_size=family_size,
+        n_children=n_children, owns_realty=owns_realty, owns_car=owns_car,
+        ext_score_1=ext_score_1, ext_score_2=ext_score_2, ext_score_3=ext_score_3,
+        bureau_count=bureau_count, bureau_active=bureau_active,
+        total_enquiries=total_enquiries, doc_count=doc_count, has_children=n_children>0,
+    )
+
+    if predict_btn or True:  # Show demo on load
+        with st.spinner("Running ML pipeline..."):
+            df_feat = build_single_applicant_features(inputs, pca, sbert)
+            
+            if model_loaded:
+                risk_score = predict_risk(df_feat, models, feature_cols)
+            else:
+                # Demo mode — compute heuristic score
+                risk_score = float(np.clip(
+                    0.9 - 0.4*np.mean([ext_score_1,ext_score_2,ext_score_3])
+                    - 0.1*(employment_yrs/40)
+                    + 0.15*(credit_amount/income if income>0 else 0.5)
+                    + np.random.normal(0, 0.02),
+                    0.01, 0.99
+                ))
+        
+        label, css_class, color, icon = risk_band(risk_score)
+        credit_score = int(300 + (1 - risk_score) * 550)  # map to 300-850 range
+        
+        # ── KPI Row ───────────────────────────────────────────────────────────
+        col1.metric("Default Probability", f"{risk_score*100:.1f}%", delta=f"{(risk_score-0.5)*100:+.1f}% vs avg")
+        col2.metric("Alt Credit Score",    f"{credit_score}", delta=None)
+        col3.metric("Risk Band",           f"{icon} {label}",  delta=None)
+        
+        st.divider()
+        
+        # ── Risk Card ─────────────────────────────────────────────────────────
+        st.markdown(f'<div class="{css_class}"><b>{icon} Risk Assessment: {label}</b><br>'
+                    f'Default probability: <b>{risk_score*100:.1f}%</b> | '
+                    f'Alternative credit score: <b>{credit_score}/850</b></div>',
+                    unsafe_allow_html=True)
+        
+        st.divider()
+        
+        # ── Tabs ──────────────────────────────────────────────────────────────
+        tab1, tab2, tab3, tab4 = st.tabs(["📊 Score Breakdown", "🔍 SHAP Explainability", "📉 Drift Simulation", "📋 Feature Profile"])
+        
+        with tab1:
+            c1, c2 = st.columns(2)
+            
+            # Gauge chart
+            fig_gauge = go.Figure(go.Indicator(
+                mode="gauge+number+delta",
+                value=credit_score,
+                delta={"reference": 650, "valueformat": ".0f"},
+                title={"text": "Alternative Credit Score", "font": {"size": 18}},
+                gauge={
+                    "axis":     {"range": [300, 850]},
+                    "bar":      {"color": color},
+                    "steps":    [
+                        {"range": [300, 550], "color": "#FFEBEE"},
+                        {"range": [550, 650], "color": "#FFF8E1"},
+                        {"range": [650, 750], "color": "#E8F5E9"},
+                        {"range": [750, 850], "color": "#C8E6C9"},
+                    ],
+                    "threshold": {"line": {"color": "red", "width": 4}, "thickness": 0.75, "value": 650},
+                }
+            ))
+            fig_gauge.update_layout(height=280, margin=dict(t=30, b=10))
+            c1.plotly_chart(fig_gauge, use_container_width=True)
+            
+            # Risk factor radar
+            categories = ["External Scores", "Income Stability", "Credit Utilisation", "Payment Behaviour", "Alt Data"]
+            ext_val    = np.mean([ext_score_1, ext_score_2, ext_score_3])
+            values = [
+                ext_val,
+                min(employment_yrs / 20, 1.0),
+                max(0, 1 - credit_amount / (income + 1) / 3),
+                ext_val * 0.9,
+                min(doc_count / 10, 1.0),
+            ]
+            
+            fig_radar = go.Figure(go.Scatterpolar(
+                r=values + [values[0]],
+                theta=categories + [categories[0]],
+                fill="toself", fillcolor=f"rgba{tuple(int(color.lstrip('#')[i:i+2],16) for i in (0,2,4)) + (0.2,)}",
+                line=dict(color=color, width=2),
+                name="Applicant"
+            ))
+            fig_radar.update_layout(
+                polar=dict(radialaxis=dict(visible=True, range=[0,1])),
+                height=280, margin=dict(t=30, b=10),
+                title="Risk Factor Radar"
+            )
+            c2.plotly_chart(fig_radar, use_container_width=True)
+        
+        with tab2:
+            if model_loaded and models:
+                st.markdown("#### SHAP Feature Attribution")
+                st.info("SHAP values show how each feature pushes the default probability up ↑ or down ↓")
+                
+                sv, base_val, df_aligned = get_shap_values(models[0], df_feat.copy(), feature_cols)
+                
+                # Sort by absolute SHAP
+                shap_df = pd.DataFrame({
+                    "Feature": df_aligned.columns,
+                    "SHAP":    sv[0],
+                    "Value":   df_aligned.iloc[0].values,
+                }).sort_values("SHAP", key=abs, ascending=False).head(15)
+                
+                colors = ["#F44336" if v > 0 else "#4CAF50" for v in shap_df["SHAP"]]
+                fig_shap = go.Figure(go.Bar(
+                    x=shap_df["SHAP"], y=shap_df["Feature"],
+                    orientation="h", marker_color=colors,
+                    text=[f"{v:+.4f}" for v in shap_df["SHAP"]], textposition="outside",
+                ))
+                fig_shap.update_layout(
+                    title="Top 15 SHAP Feature Contributions (Red = Increases Risk, Green = Decreases)",
+                    xaxis_title="SHAP Value", height=500,
+                    margin=dict(l=150)
+                )
+                st.plotly_chart(fig_shap, use_container_width=True)
+            else:
+                st.warning("⚠️ Load trained models to see SHAP explanations.")
+                # Show mock
+                mock_features = ["EXT_SOURCE_MEAN","CREDIT_INCOME_RATIO","AGE_YEARS","EMPLOYMENT_YEARS","BUREAU_COUNT",
+                                 "EXT_SOURCE_3","NLP_EMB_0","ANNUITY_INCOME_RATIO","EXT_SOURCE_1","TOTAL_ENQUIRIES"]
+                mock_shap     = np.array([-0.35, 0.28, -0.18, -0.12, 0.09, -0.22, -0.08, 0.15, -0.11, 0.06])
+                colors = ["#F44336" if v > 0 else "#4CAF50" for v in mock_shap]
+                fig_mock = go.Figure(go.Bar(
+                    x=mock_shap, y=mock_features, orientation="h",
+                    marker_color=colors, text=[f"{v:+.3f}" for v in mock_shap], textposition="outside"
+                ))
+                fig_mock.update_layout(title="Demo SHAP (load models for real values)", height=400, margin=dict(l=200))
+                st.plotly_chart(fig_mock, use_container_width=True)
+        
+        with tab3:
+            st.markdown("#### Concept Drift Sensitivity Analysis")
+            st.markdown("How does this applicant's risk score change under economic shocks?")
+            
+            income_mults = np.linspace(0.2, 1.0, 9)
+            drift_scores = []
+            for mult in income_mults:
+                drift_inp = dict(inputs)
+                drift_inp["income"] = inputs["income"] * mult
+                df_d = build_single_applicant_features(drift_inp, pca, sbert)
+                if model_loaded:
+                    s = predict_risk(df_d, models, feature_cols)
+                else:
+                    s = float(np.clip(risk_score + (1-mult)*0.25, 0, 0.99))
+                drift_scores.append(s)
+            
+            fig_drift = go.Figure()
+            fig_drift.add_trace(go.Scatter(
+                x=income_mults*100, y=[s*100 for s in drift_scores],
+                mode="lines+markers", name="Default Probability",
+                line=dict(color="#F44336", width=2.5),
+                marker=dict(size=8, color=[
+                    "#4CAF50" if s < 0.15 else "#FF9800" if s < 0.4 else "#F44336"
+                    for s in drift_scores
+                ])
+            ))
+            fig_drift.add_hline(y=40, line_dash="dash", line_color="orange", annotation_text="Medium Risk Threshold")
+            fig_drift.add_hline(y=15, line_dash="dash", line_color="green",  annotation_text="Low Risk Threshold")
+            fig_drift.update_layout(
+                title="Default Probability vs Income Shock Severity",
+                xaxis_title="Remaining Income (%)", yaxis_title="Default Probability (%)",
+                height=400
+            )
+            st.plotly_chart(fig_drift, use_container_width=True)
+        
+        with tab4:
+            st.markdown("#### Applicant Feature Summary")
+            profile_data = {
+                "Feature":       ["Annual Income", "Requested Credit", "Credit/Income Ratio", "Age",
+                                  "Employment Years", "Ext Score (Mean)", "Alt Credit Score", "Family Size"],
+                "Value":         [f"₹{income:,.0f}", f"₹{credit_amount:,.0f}",
+                                  f"{credit_amount/max(income,1):.2f}x", f"{age} yrs",
+                                  f"{employment_yrs} yrs", f"{np.mean([ext_score_1,ext_score_2,ext_score_3]):.3f}",
+                                  f"{credit_score}/850", f"{family_size} members"],
+                "Status":        ["✅" if income > 200000 else "⚠️",
+                                  "✅" if credit_amount < income*3 else "⚠️",
+                                  "✅" if credit_amount/max(income,1) < 2.5 else "🚨",
+                                  "✅", "✅" if employment_yrs > 2 else "⚠️",
+                                  "✅" if np.mean([ext_score_1,ext_score_2,ext_score_3]) > 0.5 else "🚨",
+                                  "✅" if credit_score > 650 else "⚠️", "✅"],
+            }
+            st.dataframe(pd.DataFrame(profile_data), use_container_width=True, hide_index=True)
+
+    # ── Footer ────────────────────────────────────────────────────────────────
+    st.divider()
+    st.markdown(
+        "<p style='text-align:center; font-size:0.8rem; color:#999;'>"
+        "Built with LightGBM + XGBoost + Sentence-BERT + SHAP + River (ADWIN) + W&B | "
+        "Home Credit Default Risk Dataset | "
+        "For the 1.7B credit-invisible 🌍"
+        "</p>", unsafe_allow_html=True
+    )
+
+if __name__ == "__main__":
+    main()

kaggle_output/.virtual_documents/__notebook_source__.ipynb

@@ -0,0 +1,1196 @@
+get_ipython().getoutput("uv pip install \")
+lightgbm==4.3.0 \
+xgboost==2.0.3 \
+optuna \
+optuna-integration \
+shap \
+river \
+wandb \
+kaggle \
+plotly \
+seaborn \
+imbalanced-learn \
+category-encoders category_encoders==2.6.3 transformers==4.41.2 sentence-transformers==2.7.0 numpy==1.26.4 scipy==1.11.4 scikit-learn==1.4.2
+
+
+import transformers, sentence_transformers
+print(transformers.__version__)
+print(sentence_transformers.__version__)
+
+
+import os, gc, warnings, json, re
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import seaborn as sn
+import plotly.express as px
+import plotly.graph_objects as go
+warnings.filterwarnings("ignore")
+
+# Core ML
+import lightgbm as lgb
+import xgboost as xgb
+from sklearn.model_selection import StratifiedKFold, cross_val_score
+from sklearn.preprocessing import LabelEncoder, StandardScaler, RobustScaler
+from sklearn.pipeline import Pipeline
+from sklearn.metrics import roc_auc_score, classification_report, confusion_matrix
+from sklearn.impute import SimpleImputer
+from imblearn.over_sampling import SMOTE
+import category_encoders as ce
+import optuna
+from optuna.integration import LightGBMPruningCallback
+optuna.logging.set_verbosity(optuna.logging.WARNING)
+import shap
+from river import stream, drift, linear_model, preprocessing as river_preprocessing, metrics as river_metrics, ensemble as river_ensemble
+import wandb
+
+from sentence_transformers import SentenceTransformer
+import transformers
+
+
+wandb.login(key=os.environ.get("WANDB_API_KEY", "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"))
+
+
+
+
+from dataclasses import dataclass, field
+from typing import List
+
+@dataclass
+class CFG:
+    # Paths
+    DATA_DIR: str = "/kaggle/input/competitions/home-credit-default-risk"
+    OUTPUT_DIR: str = "/kaggle/working"
+    MODEL_DIR: str = "/kaggle/working/models"
+    
+    # Experiment
+    EXPERIMENT_NAME: str = "credit-invisibility"
+    SEED: int = 42
+    N_FOLDS: int = 5
+    
+    # Training
+    LGBM_N_ITER: int = 2000
+    XGB_N_ITER: int = 1500
+    EARLY_STOPPING: int = 100
+    OPTUNA_TRIALS: int = 80
+    
+    # NLP
+    SBERT_MODEL: str = "all-MiniLM-L6-v2"   # lightweight, fast
+    NLP_DIM_REDUCTION: int = 32              # PCA to 32 dims
+    
+    # Drift simulation
+    DRIFT_FRACTION: float = 0.15
+    DRIFT_INCOME_MULTIPLIER: float = 0.4
+    
+    # W&B
+    WANDB_PROJECT: str = "credit-invisibility"
+    WANDB_ENTITY: str = None  # set your username
+
+cfg = CFG()
+os.makedirs(cfg.MODEL_DIR, exist_ok=True)
+np.random.seed(cfg.SEED)
+print("✅ Config loaded:", cfg)
+
+
+def load_all_tables(data_dir: str) -> dict:
+    tables = {}
+    files = {
+        "app_train":      "application_train.csv",
+        "app_test":       "application_test.csv",
+        "bureau":         "bureau.csv",
+        "bureau_balance": "bureau_balance.csv",
+        "prev_app":       "previous_application.csv",
+        "pos_cash":       "POS_CASH_balance.csv",
+        "installments":   "installments_payments.csv",
+        "credit_card":    "credit_card_balance.csv",
+    }
+    for key, fname in files.items():
+        path = os.path.join(data_dir, fname)
+        if os.path.exists(path):
+            tables[key] = pd.read_csv(path)
+            print(f"  ✅ {key:20s} → {tables[key].shape}")
+        else:
+            print(f"    {key:20s} → NOT FOUND")
+    return tables
+
+print("📂 Loading all Home Credit tables...")
+tables = load_all_tables(cfg.DATA_DIR)
+
+# Quick snapshot
+train = tables["app_train"]
+test  = tables["app_test"]
+
+print(f"\n Train shape: {train.shape}")
+print(f" Test shape:  {test.shape}")
+print(f"\n Target distribution:\n{train['TARGET'].value_counts(normalize=True).round(3)}")
+print(f"\n  Class imbalance ratio: {train['TARGET'].value_counts()[0]/train['TARGET'].value_counts()[1]:.1f}:1")
+
+
+fig, axes = plt.subplots(2, 3, figsize=(18, 10))
+fig.suptitle("Home Credit — Exploratory Data Analysis", fontsize=16, fontweight="bold")
+
+# 1. Target distribution
+axes[0,0].pie(
+    train["TARGET"].value_counts(),
+    labels=["Non-Default (0)", "Default (1)"],
+    autopct="%1.1f%%", colors=["#2196F3","#F44336"], startangle=90
+)
+axes[0,0].set_title("Target Distribution")
+
+# 2. Income distribution by target
+for t, color in zip([0, 1], ["#2196F3", "#F44336"]):
+    subset = train[train["TARGET"] == t]["AMT_INCOME_TOTAL"].clip(0, 500000)
+    axes[0,1].hist(subset, bins=50, alpha=0.6, label=f"Target={t}", color=color)
+axes[0,1].set_title("Income Distribution by Target")
+axes[0,1].legend()
+axes[0,1].set_xlabel("Annual Income")
+
+# 3. Credit amount by target
+for t, color in zip([0, 1], ["#2196F3", "#F44336"]):
+    subset = train[train["TARGET"] == t]["AMT_CREDIT"].clip(0, 2000000)
+    axes[0,2].hist(subset, bins=50, alpha=0.6, label=f"Target={t}", color=color)
+axes[0,2].set_title("Credit Amount by Target")
+axes[0,2].legend()
+
+# 4. Missing value heatmap (top 20)
+missing = (train.isnull().sum() / len(train) * 100).sort_values(ascending=False).head(20)
+axes[1,0].barh(missing.index, missing.values, color="#FF9800")
+axes[1,0].set_title("Top 20 Missing Value %")
+axes[1,0].set_xlabel("Missing %")
+
+# 5. Age distribution
+age_years = train["DAYS_BIRTH"].abs() / 365
+for t, color in zip([0, 1], ["#2196F3", "#F44336"]):
+    subset = age_years[train["TARGET"] == t]
+    axes[1,1].hist(subset, bins=40, alpha=0.6, label=f"Target={t}", color=color)
+axes[1,1].set_title("Age Distribution by Target")
+axes[1,1].set_xlabel("Age (years)")
+axes[1,1].legend()
+
+# 6. Occupation type default rate
+occ_default = train.groupby("OCCUPATION_TYPE")["TARGET"].mean().sort_values(ascending=False)
+axes[1,2].barh(occ_default.index, occ_default.values, color="#9C27B0")
+axes[1,2].set_title("Default Rate by Occupation")
+axes[1,2].set_xlabel("Default Rate")
+
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/eda_overview.png", dpi=150, bbox_inches="tight")
+plt.show()
+print("✅ EDA plots saved")
+
+
+def engineer_bureau_features(bureau: pd.DataFrame, bureau_balance: pd.DataFrame) -> pd.DataFrame:
+    """Extract rich signals from external credit bureau data."""
+    
+    # Bureau balance — rolling DPD (days past due) features
+    bb_agg = bureau_balance.groupby("SK_ID_BUREAU").agg(
+        STATUS_WORST  = ("STATUS", lambda x: x.map({"C":0,"0":0,"1":1,"2":2,"3":3,"4":4,"5":5,"X":0}).max()),
+        STATUS_MEAN   = ("STATUS", lambda x: x.map({"C":0,"0":0,"1":1,"2":2,"3":3,"4":4,"5":5,"X":0}).mean()),
+        MONTHS_COUNT  = ("MONTHS_BALANCE", "count"),
+    ).reset_index()
+    
+    bureau = bureau.merge(bb_agg, on="SK_ID_BUREAU", how="left")
+    
+    # Active vs closed credit
+    bureau["CREDIT_ACTIVE_BINARY"] = (bureau["CREDIT_ACTIVE"] == "Active").astype(int)
+    bureau["DAYS_CREDIT_ENDDATE"]  = bureau["DAYS_CREDIT_ENDDATE"].clip(-3000, 3000)
+    bureau["DEBT_CREDIT_RATIO"]    = bureau["AMT_CREDIT_SUM_DEBT"] / (bureau["AMT_CREDIT_SUM"] + 1)
+    bureau["CREDIT_UTIL_RATE"]     = bureau["AMT_CREDIT_SUM_OVERDUE"] / (bureau["AMT_CREDIT_SUM"] + 1)
+    
+    aggregations = {
+        "DAYS_CREDIT":            ["mean","min","max","std"],
+        "CREDIT_DAY_OVERDUE":     ["mean","max","sum"],
+        "DAYS_CREDIT_ENDDATE":    ["mean","min","max"],
+        "AMT_CREDIT_SUM":         ["mean","max","sum"],
+        "AMT_CREDIT_SUM_DEBT":    ["mean","max","sum"],
+        "AMT_CREDIT_SUM_OVERDUE": ["mean","max","sum"],
+        "DEBT_CREDIT_RATIO":      ["mean","max"],
+        "CREDIT_UTIL_RATE":       ["mean","max"],
+        "CREDIT_ACTIVE_BINARY":   ["mean","sum"],
+        "STATUS_WORST":           ["mean","max"],
+        "STATUS_MEAN":            ["mean"],
+        "MONTHS_COUNT":           ["mean","sum"],
+        "CNT_CREDIT_PROLONG":     ["sum","mean"],
+    }
+    
+    bureau_agg = bureau.groupby("SK_ID_CURR").agg(aggregations)
+    bureau_agg.columns = ["BUREAU_" + "_".join(col).upper() for col in bureau_agg.columns]
+    
+    # Count of bureau records (signal for credit footprint)
+    bureau_agg["BUREAU_COUNT"] = bureau.groupby("SK_ID_CURR").size()
+    bureau_agg["BUREAU_ACTIVE_COUNT"] = bureau.groupby("SK_ID_CURR")["CREDIT_ACTIVE_BINARY"].sum()
+    
+    return bureau_agg.reset_index()
+
+bureau_features = engineer_bureau_features(tables["bureau"], tables["bureau_balance"])
+print(f"✅ Bureau features: {bureau_features.shape}")
+
+
+def engineer_prev_app_features(prev: pd.DataFrame) -> pd.DataFrame:
+    prev["APP_CREDIT_RATIO"]       = prev["AMT_APPLICATION"] / (prev["AMT_CREDIT"] + 1)
+    prev["DOWN_PAYMENT_RATIO"]     = prev["AMT_DOWN_PAYMENT"] / (prev["AMT_CREDIT"] + 1)
+    prev["ANNUITY_CREDIT_RATIO"]   = prev["AMT_ANNUITY"] / (prev["AMT_CREDIT"] + 1)
+    prev["APPROVED"]               = (prev["NAME_CONTRACT_STATUS"] == "Approved").astype(int)
+    prev["REFUSED"]                = (prev["NAME_CONTRACT_STATUS"] == "Refused").astype(int)
+    
+    agg = prev.groupby("SK_ID_CURR").agg(
+        PREV_COUNT                = ("SK_ID_PREV",            "count"),
+        PREV_APPROVED_RATE        = ("APPROVED",              "mean"),
+        PREV_REFUSED_RATE         = ("REFUSED",               "mean"),
+        PREV_APP_CREDIT_RATIO_MEAN= ("APP_CREDIT_RATIO",      "mean"),
+        PREV_ANNUITY_MEAN         = ("AMT_ANNUITY",           "mean"),
+        PREV_CREDIT_MEAN          = ("AMT_CREDIT",            "mean"),
+        PREV_DAYS_DECISION_MEAN   = ("DAYS_DECISION",         "mean"),
+        PREV_DAYS_DECISION_MIN    = ("DAYS_DECISION",         "min"),
+        PREV_GOODS_PRICE_MEAN     = ("AMT_GOODS_PRICE",       "mean"),
+    ).reset_index()
+    agg.columns = ["SK_ID_CURR"] + ["PREV_" + c if not c.startswith("PREV") else c for c in agg.columns[1:]]
+    return agg
+
+def engineer_installments_features(inst: pd.DataFrame) -> pd.DataFrame:
+    inst["PAYMENT_DIFF"]     = inst["AMT_INSTALMENT"] - inst["AMT_PAYMENT"]
+    inst["DAYS_ENTRY_DIFF"]  = inst["DAYS_INSTALMENT"] - inst["DAYS_ENTRY_PAYMENT"]
+    inst["LATE_PAYMENT"]     = (inst["DAYS_ENTRY_DIFF"] > 0).astype(int)
+    inst["SHORT_PAYMENT"]    = (inst["PAYMENT_DIFF"] > 0).astype(int)
+    
+    agg = inst.groupby("SK_ID_CURR").agg(
+        INST_PAYMENT_DIFF_MEAN  = ("PAYMENT_DIFF",    "mean"),
+        INST_PAYMENT_DIFF_MAX   = ("PAYMENT_DIFF",    "max"),
+        INST_DAYS_ENTRY_DIFF_MEAN=("DAYS_ENTRY_DIFF", "mean"),
+        INST_LATE_PAYMENT_RATE  = ("LATE_PAYMENT",    "mean"),
+        INST_SHORT_PAYMENT_RATE = ("SHORT_PAYMENT",   "mean"),
+        INST_COUNT              = ("SK_ID_PREV",      "count"),
+    ).reset_index()
+    return agg
+
+def engineer_pos_cash_features(pos: pd.DataFrame) -> pd.DataFrame:
+    pos["DPD_BINARY"] = (pos["SK_DPD"] > 0).astype(int)
+    agg = pos.groupby("SK_ID_CURR").agg(
+        POS_MONTHS_COUNT    = ("MONTHS_BALANCE",  "count"),
+        POS_SK_DPD_MEAN     = ("SK_DPD",          "mean"),
+        POS_SK_DPD_MAX      = ("SK_DPD",          "max"),
+        POS_DPD_RATE        = ("DPD_BINARY",       "mean"),
+        POS_CNT_INSTALMENT_MEAN=("CNT_INSTALMENT", "mean"),
+    ).reset_index()
+    return agg
+
+def engineer_credit_card_features(cc: pd.DataFrame) -> pd.DataFrame:
+    cc["UTIL_RATE"]   = cc["AMT_BALANCE"] / (cc["AMT_CREDIT_LIMIT_ACTUAL"] + 1)
+    cc["DRAWING_RATE"]= cc["AMT_DRAWINGS_CURRENT"] / (cc["AMT_CREDIT_LIMIT_ACTUAL"] + 1)
+    agg = cc.groupby("SK_ID_CURR").agg(
+        CC_UTIL_RATE_MEAN   = ("UTIL_RATE",       "mean"),
+        CC_UTIL_RATE_MAX    = ("UTIL_RATE",        "max"),
+        CC_DRAWING_RATE_MEAN= ("DRAWING_RATE",     "mean"),
+        CC_AMT_BALANCE_MEAN = ("AMT_BALANCE",      "mean"),
+        CC_COUNT            = ("SK_ID_PREV",        "count"),
+        CC_DPD_MEAN         = ("SK_DPD",           "mean"),
+    ).reset_index()
+    return agg
+
+prev_features  = engineer_prev_app_features(tables["prev_app"])
+inst_features  = engineer_installments_features(tables["installments"])
+pos_features   = engineer_pos_cash_features(tables["pos_cash"])
+cc_features    = engineer_credit_card_features(tables["credit_card"])
+
+print(f"✅ Prev app features:   {prev_features.shape}")
+print(f"✅ Installments features:{inst_features.shape}")
+print(f"✅ POS Cash features:   {pos_features.shape}")
+print(f"✅ Credit Card features:{cc_features.shape}")
+
+
+def engineer_app_features(df: pd.DataFrame) -> pd.DataFrame:
+    df = df.copy()
+    
+    # Domain knowledge ratios
+    df["CREDIT_INCOME_RATIO"]       = df["AMT_CREDIT"] / (df["AMT_INCOME_TOTAL"] + 1)
+    df["ANNUITY_INCOME_RATIO"]      = df["AMT_ANNUITY"] / (df["AMT_INCOME_TOTAL"] + 1)
+    df["CREDIT_TERM"]               = df["AMT_ANNUITY"] / (df["AMT_CREDIT"] + 1)
+    df["GOODS_CREDIT_RATIO"]        = df["AMT_GOODS_PRICE"] / (df["AMT_CREDIT"] + 1)
+    
+    # Age and employment signals
+    df["AGE_YEARS"]                 = df["DAYS_BIRTH"].abs() / 365.25
+    df["EMPLOYMENT_YEARS"]          = df["DAYS_EMPLOYED"].apply(lambda x: abs(x)/365.25 if x < 0 else 0)
+    df["EMPLOYED_RATIO"]            = df["EMPLOYMENT_YEARS"] / (df["AGE_YEARS"] + 1)
+    df["CREDIT_TO_AGE"]             = df["AMT_CREDIT"] / (df["AGE_YEARS"] + 1)
+    
+    # Family / social signals
+    df["INCOME_PER_PERSON"]         = df["AMT_INCOME_TOTAL"] / (df["CNT_FAM_MEMBERS"] + 1)
+    df["CHILDREN_RATIO"]            = df["CNT_CHILDREN"] / (df["CNT_FAM_MEMBERS"] + 1)
+    
+    # External scores (most predictive in Home Credit baseline)
+    df["EXT_SOURCE_MEAN"]           = df[["EXT_SOURCE_1","EXT_SOURCE_2","EXT_SOURCE_3"]].mean(axis=1)
+    df["EXT_SOURCE_MIN"]            = df[["EXT_SOURCE_1","EXT_SOURCE_2","EXT_SOURCE_3"]].min(axis=1)
+    df["EXT_SOURCE_PROD"]           = df["EXT_SOURCE_1"] * df["EXT_SOURCE_2"] * df["EXT_SOURCE_3"]
+    df["EXT_SOURCE_STD"]            = df[["EXT_SOURCE_1","EXT_SOURCE_2","EXT_SOURCE_3"]].std(axis=1)
+    df["EXT1_EXT2_INTERACTION"]     = df["EXT_SOURCE_1"] * df["EXT_SOURCE_2"]
+    df["EXT2_EXT3_INTERACTION"]     = df["EXT_SOURCE_2"] * df["EXT_SOURCE_3"]
+    df["EXT_CREDIT_RATIO"]          = df["EXT_SOURCE_MEAN"] * df["CREDIT_INCOME_RATIO"]
+    
+    # Document flags — count missing documents
+    doc_cols = [c for c in df.columns if "FLAG_DOCUMENT" in c]
+    df["DOCUMENT_COUNT"]            = df[doc_cols].sum(axis=1)
+    
+    # Enquiry signals (loan shopping behavior)
+    enq_cols = [c for c in df.columns if "AMT_REQ_CREDIT_BUREAU" in c]
+    df["TOTAL_ENQUIRIES"]           = df[enq_cols].sum(axis=1)
+    df["RECENT_ENQUIRY_RATIO"]      = df.get("AMT_REQ_CREDIT_BUREAU_WEEK", pd.Series(0, index=df.index)) / (df["TOTAL_ENQUIRIES"] + 1)
+    
+    # Car & realty
+    df["HAS_CAR_REALTY"]            = ((df["FLAG_OWN_CAR"] == "Y") & (df["FLAG_OWN_REALTY"] == "Y")).astype(int)
+    
+    # Days registration relative to application
+    df["DAYS_REGISTRATION_RATIO"]   = df["DAYS_REGISTRATION"] / (df["DAYS_BIRTH"] + 1)
+    
+    # Label encode categoricals
+    cat_cols = df.select_dtypes("object").columns.tolist()
+    le = LabelEncoder()
+    for col in cat_cols:
+        df[col] = df[col].fillna("Unknown")
+        df[col] = le.fit_transform(df[col].astype(str))
+    
+    return df
+
+train_eng = engineer_app_features(train)
+test_eng  = engineer_app_features(test)
+print(f"✅ Engineered features: {train_eng.shape}")
+
+
+def generate_financial_descriptions(df: pd.DataFrame) -> list:
+    """
+    Synthesize a financial narrative per applicant from tabular data.
+    This simulates what a financial literacy assessment text would contain.
+    """
+    descriptions = []
+    for _, row in df.iterrows():
+        age = abs(row.get("DAYS_BIRTH", -365*35)) / 365
+        income = row.get("AMT_INCOME_TOTAL", 100000)
+        credit = row.get("AMT_CREDIT", 200000)
+        ext_score = row.get("EXT_SOURCE_MEAN", 0.5)
+        emp_years = max(0, -row.get("DAYS_EMPLOYED", -5*365)) / 365
+        
+        # Map external score to financial literacy level
+        if ext_score > 0.65:
+            literacy = "demonstrates strong financial planning habits and consistently pays bills on time"
+        elif ext_score > 0.45:
+            literacy = "shows moderate financial awareness with occasional late payments"
+        else:
+            literacy = "has limited financial experience and irregular payment patterns"
+        
+        desc = (
+            f"Applicant aged {age:.0f} years with annual income of {income:.0f} currency units. "
+            f"Requesting credit of {credit:.0f} for personal needs. "
+            f"Employed for {emp_years:.1f} years in current position. "
+            f"Client {literacy}. "
+            f"External credit assessment score: {ext_score:.2f}. "
+            f"{'Owns property which serves as collateral.' if row.get('FLAG_OWN_REALTY', 0) else 'No property ownership.'} "
+            f"{'Has dependents in household.' if row.get('CNT_CHILDREN', 0) > 0 else 'No children.'}"
+        )
+        descriptions.append(desc)
+    return descriptions
+
+print("🔤 Generating financial narratives...")
+train_texts = generate_financial_descriptions(train_eng)
+test_texts  = generate_financial_descriptions(test_eng)
+print(f"  Sample: {train_texts[0][:120]}...")
+
+# Embed with Sentence-BERT
+print("\n🤖 Loading SBERT model...")
+sbert = SentenceTransformer(cfg.SBERT_MODEL)
+
+print("  Encoding train texts (batch)...")
+train_embeddings = sbert.encode(
+    train_texts, batch_size=512, show_progress_bar=True,
+    normalize_embeddings=True, convert_to_numpy=True
+)
+
+print("  Encoding test texts (batch)...")
+test_embeddings = sbert.encode(
+    test_texts, batch_size=512, show_progress_bar=True,
+    normalize_embeddings=True, convert_to_numpy=True
+)
+
+print(f"\n✅ Embeddings shape: {train_embeddings.shape}")
+
+# Reduce dims with PCA
+from sklearn.decomposition import PCA
+pca = PCA(n_components=cfg.NLP_DIM_REDUCTION, random_state=cfg.SEED)
+train_emb_reduced = pca.fit_transform(train_embeddings)
+test_emb_reduced  = pca.transform(test_embeddings)
+
+print(f"✅ After PCA: {train_emb_reduced.shape} | Explained variance: {pca.explained_variance_ratio_.sum():.3f}")
+
+# Create DataFrame
+emb_cols = [f"NLP_EMB_{i}" for i in range(cfg.NLP_DIM_REDUCTION)]
+train_nlp_df = pd.DataFrame(train_emb_reduced, columns=emb_cols, index=train_eng.index)
+test_nlp_df  = pd.DataFrame(test_emb_reduced,  columns=emb_cols, index=test_eng.index)
+
+del sbert, train_embeddings, test_embeddings; gc.collect()
+print("✅ NLP features ready")
+
+
+def merge_all_features(app_df, bureau_feat, prev_feat, inst_feat, pos_feat, cc_feat, nlp_feat):
+    df = app_df.copy()
+    df = df.merge(bureau_feat, on="SK_ID_CURR", how="left")
+    df = df.merge(prev_feat,   on="SK_ID_CURR", how="left")
+    df = df.merge(inst_feat,   on="SK_ID_CURR", how="left")
+    df = df.merge(pos_feat,    on="SK_ID_CURR", how="left")
+    df = df.merge(cc_feat,     on="SK_ID_CURR", how="left")
+    
+    # NLP features (index-aligned)
+    nlp_feat_reset = nlp_feat.reset_index(drop=True)
+    df = df.reset_index(drop=True)
+    df = pd.concat([df, nlp_feat_reset], axis=1)
+    
+    return df
+
+print("🔗 Merging all feature tables...")
+train_full = merge_all_features(train_eng, bureau_features, prev_features, inst_features, pos_features, cc_features, train_nlp_df)
+test_full  = merge_all_features(test_eng,  bureau_features, prev_features, inst_features, pos_features, cc_features, test_nlp_df)
+
+print(f"✅ Final train shape: {train_full.shape}")
+print(f"✅ Final test shape:  {test_full.shape}")
+
+# Separate target + features
+TARGET     = "TARGET"
+DROP_COLS  = ["TARGET", "SK_ID_CURR"]
+FEATURE_COLS = [c for c in train_full.columns if c not in DROP_COLS]
+
+X = train_full[FEATURE_COLS]
+y = train_full[TARGET]
+X_test_final = test_full[FEATURE_COLS]
+
+print(f"\n✅ X: {X.shape} | y: {y.shape}")
+print(f"   NLP features: {sum(1 for c in FEATURE_COLS if 'NLP' in c)} | Tabular: {sum(1 for c in FEATURE_COLS if 'NLP' not in c)}")
+
+
+def run_lgbm_baseline(X, y, X_test, cfg, params=None):
+    """5-fold CV LightGBM with W&B logging."""
+    
+    run = wandb.init(
+        project=cfg.WANDB_PROJECT,
+        name="lgbm-baseline",
+        config=params or {},
+        tags=["baseline", "lgbm"]
+    )
+    
+    default_params = {
+        "objective":      "binary",
+        "metric":         "auc",
+        "boosting_type":  "gbdt",
+        "num_leaves":     127,
+        "learning_rate":  0.05,
+        "feature_fraction": 0.85,
+        "bagging_fraction": 0.85,
+        "bagging_freq":   5,
+        "min_child_samples": 20,
+        "reg_alpha":      0.1,
+        "reg_lambda":     0.1,
+        "n_jobs":         -1,
+        "seed":           cfg.SEED,
+        "verbose":        -1,
+    }
+    if params:
+        default_params.update(params)
+    
+    skf = StratifiedKFold(n_splits=cfg.N_FOLDS, shuffle=True, random_state=cfg.SEED)
+    oof_preds = np.zeros(len(X))
+    test_preds = np.zeros(len(X_test))
+    fold_scores = []
+    models = []
+    
+    for fold, (trn_idx, val_idx) in enumerate(skf.split(X, y)):
+        X_trn, X_val = X.iloc[trn_idx], X.iloc[val_idx]
+        y_trn, y_val = y.iloc[trn_idx], y.iloc[val_idx]
+        y_trn = y_trn.astype(np.float32)
+        y_val = y_val.astype(np.float32)
+        
+        #dtrain = lgb.Dataset(X_trn, label=y_trn)
+        #dval   = lgb.Dataset(X_val, label=y_val, reference=dtrain)
+        dtrain = lgb.Dataset(X_trn, label=np.asarray(y_trn, dtype=np.float32))
+        dval   = lgb.Dataset(X_val, label=np.asarray(y_val, dtype=np.float32), reference=dtrain)
+        
+        callbacks = [
+            lgb.early_stopping(cfg.EARLY_STOPPING, verbose=False),
+            lgb.log_evaluation(200),
+        ]
+        
+        model = lgb.train(
+            default_params,
+            dtrain,
+            num_boost_round=cfg.LGBM_N_ITER,
+            valid_sets=[dval],
+            callbacks=callbacks,
+        )
+        
+        oof_preds[val_idx] = model.predict(X_val, num_iteration=model.best_iteration)
+        test_preds += model.predict(X_test, num_iteration=model.best_iteration) / cfg.N_FOLDS
+        
+        score = roc_auc_score(y_val, oof_preds[val_idx])
+        fold_scores.append(score)
+        models.append(model)
+        
+        # W&B logging per fold
+        wandb.log({f"fold_{fold+1}_auc": score, "fold": fold+1})
+        print(f"  Fold {fold+1} | AUC: {score:.5f} | Best iter: {model.best_iteration}")
+    
+    oof_auc = roc_auc_score(y, oof_preds)
+    print(f"\n🏆 OOF AUC: {oof_auc:.5f} ± {np.std(fold_scores):.5f}")
+    
+    # Log final metrics to W&B
+    wandb.log({
+        "oof_auc":      oof_auc,
+        "fold_std":     np.std(fold_scores),
+        "n_features":   X.shape[1],
+        "n_train":      len(X),
+    })
+    
+    # Feature importance
+    fi = pd.DataFrame({
+        "feature":    X.columns,
+        "importance": np.mean([m.feature_importance("gain") for m in models], axis=0)
+    }).sort_values("importance", ascending=False)
+    
+    # Log top-20 feature importance table
+    wandb.log({"feature_importance": wandb.Table(dataframe=fi.head(20))})
+    
+    run.finish()
+    return models, oof_preds, test_preds, oof_auc, fi
+
+print("🚀 Running LightGBM Baseline...")
+lgbm_models, lgbm_oof, lgbm_test, lgbm_auc, feat_imp = run_lgbm_baseline(X, y, X_test_final, cfg)
+
+
+def objective_lgbm(trial, X, y, cfg):
+    params = {
+        "objective": "binary",
+        "metric": "auc",
+        "verbosity": -1,
+    
+        # 🔥 GPU SETTINGS
+        "device": "gpu",
+        "gpu_platform_id": 0,
+        "gpu_device_id": 0,
+    
+        # ⚡ Important for GPU stability
+        "max_bin": 255,
+        "gpu_use_dp": False,
+    
+        # Your search space
+        "boosting_type": trial.suggest_categorical("boosting_type", ["gbdt", "dart"]),
+        "num_leaves": trial.suggest_int("num_leaves", 31, 128),
+        "learning_rate": trial.suggest_float("learning_rate", 0.01, 0.1, log=True),
+        "feature_fraction": trial.suggest_float("feature_fraction", 0.6, 1.0),
+        "bagging_fraction": trial.suggest_float("bagging_fraction", 0.6, 1.0),
+        "bagging_freq": trial.suggest_int("bagging_freq", 1, 7),
+        "min_child_samples": trial.suggest_int("min_child_samples", 5, 100),
+        "reg_alpha": trial.suggest_float("reg_alpha", 1e-4, 10.0, log=True),
+        "reg_lambda": trial.suggest_float("reg_lambda", 1e-4, 10.0, log=True),
+        "max_depth": trial.suggest_int("max_depth", 4, 10),
+        "min_gain_to_split": trial.suggest_float("min_gain_to_split", 0.0, 0.5),
+    
+        "n_jobs": -1,
+        "seed": cfg.SEED,
+    }
+    
+    params["force_col_wise"] = True
+    skf = StratifiedKFold(n_splits=3, shuffle=True, random_state=cfg.SEED)
+    scores = []
+    
+    for trn_idx, val_idx in skf.split(X, y):
+        X_trn, X_val = X.iloc[trn_idx], X.iloc[val_idx]
+        y_trn, y_val = y.iloc[trn_idx], y.iloc[val_idx]
+
+
+        y_trn_np = np.asarray(y_trn, dtype=np.float32)
+        y_val_np = np.asarray(y_val, dtype=np.float32)
+        
+        dtrain = lgb.Dataset(X_trn, label=y_trn_np)
+        dval   = lgb.Dataset(X_val, label=y_val_np, reference=dtrain)
+
+
+        
+        #dtrain = lgb.Dataset(X_trn, label=y_trn)
+        #dval   = lgb.Dataset(X_val, label=y_val, reference=dtrain)
+        
+        pruning_cb = LightGBMPruningCallback(trial, "auc")
+        
+        model = lgb.train(
+            params, dtrain,
+            num_boost_round=1000,
+            valid_sets=[dval],
+            callbacks=[
+                lgb.early_stopping(50, verbose=False),
+                lgb.log_evaluation(-1),
+                pruning_cb,
+            ]
+        )
+        preds = model.predict(X_val, num_iteration=model.best_iteration)
+        scores.append(roc_auc_score(y_val, preds))
+    
+    return np.mean(scores)
+
+print("🔍 Running Optuna HPO for LightGBM...")
+print(f"   Trials: {cfg.OPTUNA_TRIALS}")
+
+sampler  = optuna.samplers.TPESampler(seed=cfg.SEED)
+pruner   = optuna.pruners.MedianPruner(n_warmup_steps=10)
+study_lgbm = optuna.create_study(
+    direction="maximize", sampler=sampler, pruner=pruner,
+    study_name="lgbm-credit-hpo"
+)
+
+study_lgbm.optimize(
+    lambda trial: objective_lgbm(trial, X, y, cfg),
+    n_trials=cfg.OPTUNA_TRIALS,
+    show_progress_bar=True,
+    n_jobs=1,
+)
+
+best_lgbm_params = study_lgbm.best_params
+best_lgbm_params.update({"objective":"binary","metric":"auc","verbosity":-1,"n_jobs":-1,"seed":cfg.SEED})
+
+print(f"\n🏆 Best LightGBM AUC: {study_lgbm.best_value:.5f}")
+print(f"   Best params: {json.dumps(best_lgbm_params, indent=2)}")
+
+
+
+
+def objective_xgb(trial, X, y, cfg):
+    params = {
+        "objective":          "binary:logistic",
+        "tree_method": "hist",
+        "device": "cuda",          # ✅ enables GPU
+        "max_bin": 256,            # ⚡ important for GPU speed
+        "eval_metric":        "auc",
+        "tree_method":        "hist",
+        "use_label_encoder":  False,
+        "learning_rate":      trial.suggest_float("learning_rate", 0.01, 0.1, log=True),
+        "max_depth":          trial.suggest_int("max_depth", 4, 10),
+        "min_child_weight":   trial.suggest_int("min_child_weight", 1, 20),
+        "subsample":          trial.suggest_float("subsample", 0.6, 1.0),
+        "colsample_bytree":   trial.suggest_float("colsample_bytree", 0.5, 1.0),
+        "gamma":              trial.suggest_float("gamma", 0.0, 2.0),
+        "reg_alpha":          trial.suggest_float("reg_alpha", 1e-4, 10.0, log=True),
+        "reg_lambda":         trial.suggest_float("reg_lambda", 1e-4, 10.0, log=True),
+        "scale_pos_weight":   trial.suggest_float("scale_pos_weight", 1.0, 10.0),
+        "seed":               cfg.SEED,
+        "n_jobs":             -1,
+    }
+    
+    skf = StratifiedKFold(n_splits=3, shuffle=True, random_state=cfg.SEED)
+    scores = []
+    
+    for trn_idx, val_idx in skf.split(X, y):
+        X_trn, X_val = X.iloc[trn_idx], X.iloc[val_idx]
+        y_trn, y_val = y.iloc[trn_idx], y.iloc[val_idx]
+        
+        dtrain = xgb.DMatrix(X_trn, label=np.asarray(y_trn, dtype=np.float32))
+        dval   = xgb.DMatrix(X_val, label=np.asarray(y_val, dtype=np.float32))
+
+        
+        model = xgb.train(
+            params, dtrain,
+            num_boost_round=1000,
+            evals=[(dval, "val")],
+            early_stopping_rounds=50,
+            verbose_eval=False,
+        )
+        preds = model.predict(dval)
+        scores.append(roc_auc_score(y_val, preds))
+    
+    return np.mean(scores)
+
+print("🔍 Running Optuna HPO for XGBoost...")
+study_xgb = optuna.create_study(
+    direction="maximize",
+    sampler=optuna.samplers.TPESampler(seed=cfg.SEED),
+    study_name="xgb-credit-hpo"
+)
+study_xgb.optimize(
+    lambda trial: objective_xgb(trial, X, y, cfg),
+    n_trials=cfg.OPTUNA_TRIALS,
+    show_progress_bar=True,
+)
+
+best_xgb_params = study_xgb.best_params
+best_xgb_params.update({"objective":"binary:logistic","eval_metric":"auc","tree_method":"hist","seed":cfg.SEED,"n_jobs":-1})
+
+print(f"\n🏆 Best XGBoost AUC: {study_xgb.best_value:.5f}")
+
+
+def train_full_ensemble(X, y, X_test, lgbm_params, xgb_params, cfg):
+
+    lgbm_params.update({
+        "device": "gpu",
+        "max_bin": 255,
+        "gpu_use_dp": False,
+        "force_col_wise": True
+    })
+
+    xgb_params.update({
+        "device": "cuda",
+        "tree_method": "hist",
+        "max_bin": 256
+    })
+
+    run = wandb.init(
+        project=cfg.WANDB_PROJECT,
+        name="ensemble-lgbm-xgb",
+        config={"lgbm": lgbm_params, "xgb": xgb_params, "n_folds": cfg.N_FOLDS},
+        tags=["ensemble", "final"]
+    )
+
+    skf = StratifiedKFold(n_splits=cfg.N_FOLDS, shuffle=True, random_state=cfg.SEED)
+
+    oof_lgbm  = np.zeros(len(X))
+    oof_xgb   = np.zeros(len(X))
+    test_lgbm = np.zeros(len(X_test))
+    test_xgb  = np.zeros(len(X_test))
+
+    lgbm_models_list = []
+    xgb_models_list  = []
+
+    for fold, (trn_idx, val_idx) in enumerate(skf.split(X, y)):
+        print(f"\n🚀 Fold {fold+1}")
+
+        X_trn, X_val = X.iloc[trn_idx], X.iloc[val_idx]
+        y_trn, y_val = y.iloc[trn_idx], y.iloc[val_idx]
+
+        y_trn_np = np.asarray(y_trn, dtype=np.float32)
+        y_val_np = np.asarray(y_val, dtype=np.float32)
+
+        dl_trn = lgb.Dataset(X_trn, label=y_trn_np)
+        dl_val = lgb.Dataset(X_val, label=y_val_np, reference=dl_trn)
+
+        lgb_model = lgb.train(
+            lgbm_params,
+            dl_trn,
+            num_boost_round=cfg.LGBM_N_ITER,
+            valid_sets=[dl_val],
+            callbacks=[
+                lgb.early_stopping(cfg.EARLY_STOPPING, verbose=False),
+                lgb.log_evaluation(100),
+            ]
+        )
+
+        lgb_val_pred = lgb_model.predict(X_val, num_iteration=lgb_model.best_iteration)
+        oof_lgbm[val_idx] = lgb_val_pred
+        test_lgbm += lgb_model.predict(X_test, num_iteration=lgb_model.best_iteration) / cfg.N_FOLDS
+        lgbm_models_list.append(lgb_model)
+
+        dx_trn = xgb.DMatrix(X_trn, label=y_trn_np)
+        dx_val = xgb.DMatrix(X_val, label=y_val_np)
+        dx_tst = xgb.DMatrix(X_test)
+
+        xgb_model = xgb.train(
+            xgb_params,
+            dx_trn,
+            num_boost_round=cfg.XGB_N_ITER,
+            evals=[(dx_val, "val")],
+            early_stopping_rounds=cfg.EARLY_STOPPING,
+            verbose_eval=False,
+        )
+
+        xgb_val_pred = xgb_model.predict(dx_val)
+        oof_xgb[val_idx] = xgb_val_pred
+        test_xgb += xgb_model.predict(dx_tst) / cfg.N_FOLDS
+        xgb_models_list.append(xgb_model)
+
+        # Fold metrics
+        auc_l = roc_auc_score(y_val, lgb_val_pred)
+        auc_x = roc_auc_score(y_val, xgb_val_pred)
+
+        print(f"  LGBM: {auc_l:.5f} | XGB: {auc_x:.5f}")
+
+
+    print("\n🔍 Optimizing blend weights...")
+
+    best_auc = 0
+    best_w = 0.5
+
+    for w in np.arange(0.0, 1.01, 0.01):
+        blend = w * oof_lgbm + (1 - w) * oof_xgb
+        auc = roc_auc_score(y, blend)
+
+        if auc > best_auc:
+            best_auc = auc
+            best_w = w
+
+    print(f"✅ Best weight → LGBM: {best_w:.2f}, XGB: {1-best_w:.2f}")
+    print(f"🏆 Best OOF AUC: {best_auc:.5f}")
+
+    oof_blend  = best_w * oof_lgbm + (1 - best_w) * oof_xgb
+    test_blend = best_w * test_lgbm + (1 - best_w) * test_xgb
+
+    wandb.log({
+        "final_oof_auc": best_auc,
+        "lgbm_weight": best_w,
+        "xgb_weight": 1 - best_w
+    })
+
+    run.finish()
+
+    return lgbm_models_list, xgb_models_list, oof_lgbm, oof_xgb, oof_blend, test_blend
+
+
+lgbm_models_final, xgb_models_final, oof_lgbm, oof_xgb, oof_blend, test_blend = train_full_ensemble(
+    X, y, X_test_final, best_lgbm_params, best_xgb_params, cfg
+)
+
+
+print("🔍 Computing SHAP values (TreeExplainer)...")
+
+# Use the first fold's LightGBM model for SHAP analysis
+explainer = shap.TreeExplainer(lgbm_models_final[0])
+
+# Sample 2000 rows for speed
+sample_idx = np.random.choice(len(X), min(2000, len(X)), replace=False)
+X_sample   = X.iloc[sample_idx]
+shap_vals  = explainer.shap_values(X_sample)
+
+# For binary classification, lgbm returns list [neg_class, pos_class]
+if isinstance(shap_vals, list):
+    shap_vals = shap_vals[1]
+
+print(f"✅ SHAP values shape: {shap_vals.shape}")
+
+# ── 1. Beeswarm / Summary Plot ──────────────────────────────────────
+fig, ax = plt.subplots(figsize=(12, 10))
+shap.summary_plot(shap_vals, X_sample, plot_type="dot", max_display=25, show=False)
+plt.title("SHAP Beeswarm — Feature Impact on Credit Risk", fontsize=14, fontweight="bold")
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/shap_beeswarm.png", dpi=150, bbox_inches="tight")
+plt.show()
+
+# ── 2. Bar Plot (mean |SHAP|) ────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(12, 8))
+shap.summary_plot(shap_vals, X_sample, plot_type="bar", max_display=20, show=False)
+plt.title("Mean |SHAP| — Global Feature Importance", fontsize=14, fontweight="bold")
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/shap_bar.png", dpi=150, bbox_inches="tight")
+plt.show()
+
+# ── 3. Dependence Plots for top 3 features ──────────────────────────
+top3_features = pd.DataFrame({
+    "feature": X.columns,
+    "mean_shap": np.abs(shap_vals).mean(0)
+}).nlargest(3, "mean_shap")["feature"].tolist()
+
+fig, axes = plt.subplots(1, 3, figsize=(18, 5))
+for ax, feat in zip(axes, top3_features):
+    feat_idx = list(X.columns).index(feat)
+    shap.dependence_plot(feat_idx, shap_vals, X_sample, ax=ax, show=False)
+    ax.set_title(f"SHAP Dependence: {feat}", fontsize=10)
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/shap_dependence.png", dpi=150, bbox_inches="tight")
+plt.show()
+
+# ── 4. Waterfall for single applicant (most risky) ───────────────────
+most_risky_idx = np.argmax(oof_blend[sample_idx])
+expl_obj = shap.Explanation(
+    values     = shap_vals[most_risky_idx],
+    base_values= explainer.expected_value if not isinstance(explainer.expected_value, list) else explainer.expected_value[1],
+    data       = X_sample.iloc[most_risky_idx].values,
+    feature_names=X_sample.columns.tolist()
+)
+plt.figure(figsize=(14, 8))
+shap.plots.waterfall(expl_obj, max_display=15, show=False)
+plt.title("SHAP Waterfall — Most Risky Applicant", fontsize=13, fontweight="bold")
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/shap_waterfall.png", dpi=150, bbox_inches="tight")
+plt.show()
+
+print("✅ All SHAP plots saved")
+
+
+print("📉 Simulating Concept Drift...")
+
+run = wandb.init(
+    project=cfg.WANDB_PROJECT,
+    name="concept-drift-simulation",
+    tags=["drift", "simulation"]
+)
+
+# ── Get validation split (index-aligned with X, y) ────────────────────────────
+skf = StratifiedKFold(n_splits=2, shuffle=True, random_state=cfg.SEED)
+trn_idx, val_idx = list(skf.split(X, y))[0]
+
+y_val_drift = y.iloc[val_idx].values
+
+# Baseline AUC using ensemble OOF preds (already computed over full X)
+base_preds = oof_blend[val_idx]
+base_auc   = roc_auc_score(y_val_drift, base_preds)
+print(f"  Baseline AUC (no drift): {base_auc:.5f}")
+
+# Top-30 for identifying WHICH columns to perturb
+top30 = feat_imp.head(30)["feature"].tolist()
+
+# Drift scenarios
+drift_scenarios = {
+    "Baseline (No Drift)":        {"income_mult": 1.0, "emp_mask": 0.0,  "label_noise": 0.0},
+    "Mild Income Shock (-30%)":   {"income_mult": 0.7, "emp_mask": 0.05, "label_noise": 0.02},
+    "Severe Income Shock (-60%)": {"income_mult": 0.4, "emp_mask": 0.15, "label_noise": 0.05},
+    "Mass Job Loss (20%)":        {"income_mult": 0.5, "emp_mask": 0.20, "label_noise": 0.08},
+    "Full Economic Shock":        {"income_mult": 0.3, "emp_mask": 0.35, "label_noise": 0.12},
+}
+
+drift_results = []
+
+for scenario_name, drift_cfg in drift_scenarios.items():
+    # ── Start from the FULL feature matrix (234 cols) ────────────────────────
+    X_drifted = X.iloc[val_idx].copy().reset_index(drop=True)   # shape: (n_val, 234)
+
+    # Apply income shock to relevant columns (wherever they exist in full matrix)
+    income_cols = [c for c in X_drifted.columns if c in
+                   ["AMT_INCOME_TOTAL", "INCOME_PER_PERSON", "CREDIT_INCOME_RATIO",
+                    "ANNUITY_INCOME_RATIO", "EXT_CREDIT_RATIO"]]
+    for col in income_cols:
+        X_drifted[col] *= drift_cfg["income_mult"]
+
+    # Apply employment shock
+    emp_cols = [c for c in X_drifted.columns if "EMPLOY" in c or "DAYS_EMPLOYED" in c]
+    mask = np.random.random(len(X_drifted)) < drift_cfg["emp_mask"]
+    for col in emp_cols:
+        X_drifted.loc[mask, col] = 0
+
+    # Label noise
+    y_noisy = y_val_drift.copy()
+    if drift_cfg["label_noise"] > 0:
+        noise_n   = max(1, int(drift_cfg["label_noise"] * len(y_noisy)))
+        noise_idx = np.random.choice(len(y_noisy), noise_n, replace=False)
+        y_noisy[noise_idx] = 1 - y_noisy[noise_idx]
+
+    # ── Predict with the full-feature model ──────────────────────────────────
+    drifted_preds = lgbm_models_final[0].predict(X_drifted)   # 234 cols ✅
+    drifted_auc   = roc_auc_score(y_noisy, drifted_preds)
+
+    drift_results.append({
+        "scenario":    scenario_name,
+        "auc":         drifted_auc,
+        "auc_drop":    base_auc - drifted_auc,
+        "income_mult": drift_cfg["income_mult"],
+        "emp_mask":    drift_cfg["emp_mask"],
+    })
+    wandb.log({"scenario": scenario_name, "drifted_auc": drifted_auc})
+    print(f"  {scenario_name:40s} | AUC: {drifted_auc:.5f} | Drop: {base_auc - drifted_auc:+.5f}")
+
+# ── Plots ─────────────────────────────────────────────────────────────────────
+drift_df = pd.DataFrame(drift_results)
+fig, axes = plt.subplots(1, 2, figsize=(14, 5))
+
+axes[0].bar(range(len(drift_df)), drift_df["auc"],
+            color=["#2196F3","#4CAF50","#FF9800","#F44336","#9C27B0"])
+axes[0].axhline(y=0.7, color="red", linestyle="--", label="Min Acceptable AUC")
+axes[0].set_xticks(range(len(drift_df)))
+axes[0].set_xticklabels([s.split("(")[0].strip() for s in drift_df["scenario"]],
+                         rotation=20, ha="right")
+axes[0].set_ylabel("AUC")
+axes[0].set_title("Model Performance Under Drift Scenarios")
+axes[0].legend()
+
+axes[1].plot(drift_df["income_mult"], drift_df["auc"],
+             "o-", color="#F44336", linewidth=2, markersize=8)
+axes[1].set_xlabel("Income Multiplier (1.0 = no drift)")
+axes[1].set_ylabel("AUC")
+axes[1].set_title("AUC Degradation vs Income Shock")
+axes[1].fill_between(drift_df["income_mult"], drift_df["auc"], 0.5,
+                      alpha=0.15, color="#F44336")
+
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/drift_simulation.png", dpi=150, bbox_inches="tight")
+plt.show()
+
+run.finish()
+print("✅ Drift simulation complete")
+
+
+from river import (
+    drift as river_drift,
+    linear_model as river_lm,
+    preprocessing as river_pp,
+    metrics as river_metrics,
+    ensemble as river_ens,
+    tree as river_tree,
+    optim,
+)
+import time
+
+print("🌊 Initialising River online learning pipeline...")
+
+# Build River pipeline: StandardScaler → Hoeffding Adaptive Tree
+river_pipeline = river_pp.StandardScaler() | river_tree.HoeffdingAdaptiveTreeClassifier(
+    grace_period=200,
+    delta=1e-5,
+    seed=cfg.SEED,
+)
+
+# ADWIN drift detector — detects distribution shifts in the error stream
+adwin = river_drift.ADWIN(delta=0.002)
+
+# Track metrics
+river_auc      = river_metrics.ROCAUC()
+drift_points   = []
+retrain_count  = 0
+running_errors = []
+
+# Simulate streaming with drift injection
+# Use top 30 features for speed
+X_stream = X[top30].fillna(0).reset_index(drop=True)
+y_stream = y.reset_index(drop=True)
+
+# Inject drift at 60% of the way through
+DRIFT_INJECT_AT = int(len(X_stream) * 0.6)
+DRIFT_DURATION  = 5000
+
+print(f"🔄 Streaming {len(X_stream):,} samples...")
+print(f"   Drift will be injected at sample {DRIFT_INJECT_AT:,}")
+
+start = time.time()
+
+for i, (xi, yi) in enumerate(stream.iter_pandas(X_stream, y_stream)):
+    # Inject concept drift
+    if DRIFT_INJECT_AT <= i < DRIFT_INJECT_AT + DRIFT_DURATION:
+        xi["AMT_INCOME_TOTAL"] = xi.get("AMT_INCOME_TOTAL", 0) * cfg.DRIFT_INCOME_MULTIPLIER
+        yi = 1 - yi if np.random.random() < 0.12 else yi  # label noise
+    
+    # Predict → update metric → learn
+    y_prob = river_pipeline.predict_proba_one(xi)
+    p1     = y_prob.get(1, 0.5)
+    river_auc.update(yi, p1)
+    
+    error = abs(yi - p1)
+    running_errors.append(error)
+    adwin.update(error)
+    
+    if adwin.drift_detected:
+        drift_points.append(i)
+        retrain_count += 1
+        # Reset learner on drift (warm-start)
+        river_pipeline = river_pp.StandardScaler() | river_tree.HoeffdingAdaptiveTreeClassifier(
+            grace_period=50,  # faster adaptation post-drift
+            delta=1e-5,
+            seed=cfg.SEED,
+        )
+        if retrain_count <= 5:
+            print(f"  🚨 DRIFT DETECTED at sample {i:,} | Retrain #{retrain_count} | Running AUC: {river_auc.get():.4f}")
+    
+    river_pipeline.learn_one(xi, yi)
+
+elapsed = time.time() - start
+print(f"\n✅ Online learning complete in {elapsed:.1f}s")
+print(f"   Final AUC: {river_auc.get():.5f}")
+print(f"   Total drift detections: {len(drift_points)}")
+print(f"   Total retrains: {retrain_count}")
+
+# Plot error stream + drift points
+fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(14, 8))
+
+window = 500
+smoothed = pd.Series(running_errors).rolling(window).mean()
+ax1.plot(smoothed, color="#2196F3", linewidth=1, label=f"Error (rolling {window})")
+for dp in drift_points:
+    ax1.axvline(dp, color="red", linewidth=0.8, alpha=0.7)
+ax1.axvline(DRIFT_INJECT_AT, color="orange", linewidth=2, linestyle="--", label="Drift Injected")
+ax1.set_title("ADWIN Drift Detection — Error Stream", fontweight="bold")
+ax1.set_xlabel("Sample")
+ax1.set_ylabel("Prediction Error")
+ax1.legend()
+
+# Cumulative drift detections
+ax2.step(drift_points, range(1, len(drift_points)+1), color="#F44336", linewidth=2)
+ax2.axvline(DRIFT_INJECT_AT, color="orange", linewidth=2, linestyle="--", label="Drift Injected")
+ax2.set_title("Cumulative Drift Detections", fontweight="bold")
+ax2.set_xlabel("Sample Index")
+ax2.set_ylabel("Cumulative Detections")
+ax2.legend()
+
+plt.tight_layout()
+plt.savefig(f"{cfg.OUTPUT_DIR}/river_drift_detection.png", dpi=150, bbox_inches="tight")
+plt.show()
+
+
+import pickle, joblib
+
+# Save LightGBM models
+for i, model in enumerate(lgbm_models_final):
+    model.save_model(f"{cfg.MODEL_DIR}/lgbm_fold_{i+1}.txt")
+
+# Save XGBoost models
+for i, model in enumerate(xgb_models_final):
+    model.save_model(f"{cfg.MODEL_DIR}/xgb_fold_{i+1}.json")
+
+# Save PCA + config
+joblib.dump(pca, f"{cfg.MODEL_DIR}/pca.pkl")
+joblib.dump(scaler_drift, f"{cfg.MODEL_DIR}/scaler.pkl")
+
+# Save feature list
+with open(f"{cfg.MODEL_DIR}/feature_cols.json", "w") as f:
+    json.dump(FEATURE_COLS, f)
+
+print(f"✅ Models saved to {cfg.MODEL_DIR}")
+
+# Generate submission
+submission = pd.DataFrame({
+    "SK_ID_CURR": test["SK_ID_CURR"],
+    "TARGET":     test_blend
+})
+submission.to_csv(f"{cfg.OUTPUT_DIR}/submission_ensemble.csv", index=False)
+print(f"✅ Submission saved: {submission.shape}")
+print(submission.head())
+
+# W&B — final summary run
+run = wandb.init(project=cfg.WANDB_PROJECT, name="final-summary", tags=["summary"])
+wandb.log({
+    "lgbm_baseline_auc":   lgbm_auc,
+    "lgbm_optuna_best":    study_lgbm.best_value,
+    "xgb_optuna_best":     study_xgb.best_value,
+    "ensemble_oof_auc":    roc_auc_score(y, oof_blend),
+    "drift_detections":    len(drift_points),
+    "total_features":      len(FEATURE_COLS),
+    "nlp_features":        cfg.NLP_DIM_REDUCTION,
+})
+wandb.save(f"{cfg.OUTPUT_DIR}/submission_ensemble.csv")
+run.finish()
+print("✅ All done. W&B summary logged.")
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

kaggle_output/explainable-credit-risk-modeling-with-alternative.log

@@ -0,0 +1,15 @@
+[{"stream_name":"stderr","time":5.000587062,"data":"/usr/local/lib/python3.12/dist-packages/mistune.py:435: SyntaxWarning: invalid escape sequence '\\|'\n"}
+,{"stream_name":"stderr","time":5.000774502,"data":"  cells[i][c] = re.sub('\\\\\\\\\\|', '|', cell)\n"}
+,{"stream_name":"stderr","time":5.657678067,"data":"/usr/local/lib/python3.12/dist-packages/nbconvert/filters/filter_links.py:36: SyntaxWarning: invalid escape sequence '\\_'\n"}
+,{"stream_name":"stderr","time":5.657724537,"data":"  text = re.sub(r'_', '\\_', text) # Escape underscores in display text\n"}
+,{"stream_name":"stderr","time":7.211358548,"data":"[NbConvertApp] Converting notebook __notebook__.ipynb to html\n"}
+,{"stream_name":"stderr","time":9.655510019,"data":"[NbConvertApp] Support files will be in __results___files/\n"}
+,{"stream_name":"stderr","time":9.656032529,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.657319549,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.658246489,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.660020589,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.661195168999999,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.663283069,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.664512739,"data":"[NbConvertApp] Making directory __results___files\n"}
+,{"stream_name":"stderr","time":9.665452399,"data":"[NbConvertApp] Writing 558261 bytes to __results__.html\n"}
+]

kaggle_output/models/feature_cols.json

@@ -0,0 +1 @@
+["NAME_CONTRACT_TYPE", "CODE_GENDER", "FLAG_OWN_CAR", "FLAG_OWN_REALTY", "CNT_CHILDREN", "AMT_INCOME_TOTAL", "AMT_CREDIT", "AMT_ANNUITY", "AMT_GOODS_PRICE", "NAME_TYPE_SUITE", "NAME_INCOME_TYPE", "NAME_EDUCATION_TYPE", "NAME_FAMILY_STATUS", "NAME_HOUSING_TYPE", "REGION_POPULATION_RELATIVE", "DAYS_BIRTH", "DAYS_EMPLOYED", "DAYS_REGISTRATION", "DAYS_ID_PUBLISH", "OWN_CAR_AGE", "FLAG_MOBIL", "FLAG_EMP_PHONE", "FLAG_WORK_PHONE", "FLAG_CONT_MOBILE", "FLAG_PHONE", "FLAG_EMAIL", "OCCUPATION_TYPE", "CNT_FAM_MEMBERS", "REGION_RATING_CLIENT", "REGION_RATING_CLIENT_W_CITY", "WEEKDAY_APPR_PROCESS_START", "HOUR_APPR_PROCESS_START", "REG_REGION_NOT_LIVE_REGION", "REG_REGION_NOT_WORK_REGION", "LIVE_REGION_NOT_WORK_REGION", "REG_CITY_NOT_LIVE_CITY", "REG_CITY_NOT_WORK_CITY", "LIVE_CITY_NOT_WORK_CITY", "ORGANIZATION_TYPE", "EXT_SOURCE_1", "EXT_SOURCE_2", "EXT_SOURCE_3", "APARTMENTS_AVG", "BASEMENTAREA_AVG", "YEARS_BEGINEXPLUATATION_AVG", "YEARS_BUILD_AVG", "COMMONAREA_AVG", "ELEVATORS_AVG", "ENTRANCES_AVG", "FLOORSMAX_AVG", "FLOORSMIN_AVG", "LANDAREA_AVG", "LIVINGAPARTMENTS_AVG", "LIVINGAREA_AVG", "NONLIVINGAPARTMENTS_AVG", "NONLIVINGAREA_AVG", "APARTMENTS_MODE", "BASEMENTAREA_MODE", "YEARS_BEGINEXPLUATATION_MODE", "YEARS_BUILD_MODE", "COMMONAREA_MODE", "ELEVATORS_MODE", "ENTRANCES_MODE", "FLOORSMAX_MODE", "FLOORSMIN_MODE", "LANDAREA_MODE", "LIVINGAPARTMENTS_MODE", "LIVINGAREA_MODE", "NONLIVINGAPARTMENTS_MODE", "NONLIVINGAREA_MODE", "APARTMENTS_MEDI", "BASEMENTAREA_MEDI", "YEARS_BEGINEXPLUATATION_MEDI", "YEARS_BUILD_MEDI", "COMMONAREA_MEDI", "ELEVATORS_MEDI", "ENTRANCES_MEDI", "FLOORSMAX_MEDI", "FLOORSMIN_MEDI", "LANDAREA_MEDI", "LIVINGAPARTMENTS_MEDI", "LIVINGAREA_MEDI", "NONLIVINGAPARTMENTS_MEDI", "NONLIVINGAREA_MEDI", "FONDKAPREMONT_MODE", "HOUSETYPE_MODE", "TOTALAREA_MODE", "WALLSMATERIAL_MODE", "EMERGENCYSTATE_MODE", "OBS_30_CNT_SOCIAL_CIRCLE", "DEF_30_CNT_SOCIAL_CIRCLE", "OBS_60_CNT_SOCIAL_CIRCLE", "DEF_60_CNT_SOCIAL_CIRCLE", "DAYS_LAST_PHONE_CHANGE", "FLAG_DOCUMENT_2", "FLAG_DOCUMENT_3", "FLAG_DOCUMENT_4", "FLAG_DOCUMENT_5", "FLAG_DOCUMENT_6", "FLAG_DOCUMENT_7", "FLAG_DOCUMENT_8", "FLAG_DOCUMENT_9", "FLAG_DOCUMENT_10", "FLAG_DOCUMENT_11", "FLAG_DOCUMENT_12", "FLAG_DOCUMENT_13", "FLAG_DOCUMENT_14", "FLAG_DOCUMENT_15", "FLAG_DOCUMENT_16", "FLAG_DOCUMENT_17", "FLAG_DOCUMENT_18", "FLAG_DOCUMENT_19", "FLAG_DOCUMENT_20", "FLAG_DOCUMENT_21", "AMT_REQ_CREDIT_BUREAU_HOUR", "AMT_REQ_CREDIT_BUREAU_DAY", "AMT_REQ_CREDIT_BUREAU_WEEK", "AMT_REQ_CREDIT_BUREAU_MON", "AMT_REQ_CREDIT_BUREAU_QRT", "AMT_REQ_CREDIT_BUREAU_YEAR", "CREDIT_INCOME_RATIO", "ANNUITY_INCOME_RATIO", "CREDIT_TERM", "GOODS_CREDIT_RATIO", "AGE_YEARS", "EMPLOYMENT_YEARS", "EMPLOYED_RATIO", "CREDIT_TO_AGE", "INCOME_PER_PERSON", "CHILDREN_RATIO", "EXT_SOURCE_MEAN", "EXT_SOURCE_MIN", "EXT_SOURCE_PROD", "EXT_SOURCE_STD", "EXT1_EXT2_INTERACTION", "EXT2_EXT3_INTERACTION", "EXT_CREDIT_RATIO", "DOCUMENT_COUNT", "TOTAL_ENQUIRIES", "RECENT_ENQUIRY_RATIO", "HAS_CAR_REALTY", "DAYS_REGISTRATION_RATIO", "BUREAU_DAYS_CREDIT_MEAN", "BUREAU_DAYS_CREDIT_MIN", "BUREAU_DAYS_CREDIT_MAX", "BUREAU_DAYS_CREDIT_STD", "BUREAU_CREDIT_DAY_OVERDUE_MEAN", "BUREAU_CREDIT_DAY_OVERDUE_MAX", "BUREAU_CREDIT_DAY_OVERDUE_SUM", "BUREAU_DAYS_CREDIT_ENDDATE_MEAN", "BUREAU_DAYS_CREDIT_ENDDATE_MIN", "BUREAU_DAYS_CREDIT_ENDDATE_MAX", "BUREAU_AMT_CREDIT_SUM_MEAN", "BUREAU_AMT_CREDIT_SUM_MAX", "BUREAU_AMT_CREDIT_SUM_SUM", "BUREAU_AMT_CREDIT_SUM_DEBT_MEAN", "BUREAU_AMT_CREDIT_SUM_DEBT_MAX", "BUREAU_AMT_CREDIT_SUM_DEBT_SUM", "BUREAU_AMT_CREDIT_SUM_OVERDUE_MEAN", "BUREAU_AMT_CREDIT_SUM_OVERDUE_MAX", "BUREAU_AMT_CREDIT_SUM_OVERDUE_SUM", "BUREAU_DEBT_CREDIT_RATIO_MEAN", "BUREAU_DEBT_CREDIT_RATIO_MAX", "BUREAU_CREDIT_UTIL_RATE_MEAN", "BUREAU_CREDIT_UTIL_RATE_MAX", "BUREAU_CREDIT_ACTIVE_BINARY_MEAN", "BUREAU_CREDIT_ACTIVE_BINARY_SUM", "BUREAU_STATUS_WORST_MEAN", "BUREAU_STATUS_WORST_MAX", "BUREAU_STATUS_MEAN_MEAN", "BUREAU_MONTHS_COUNT_MEAN", "BUREAU_MONTHS_COUNT_SUM", "BUREAU_CNT_CREDIT_PROLONG_SUM", "BUREAU_CNT_CREDIT_PROLONG_MEAN", "BUREAU_COUNT", "BUREAU_ACTIVE_COUNT", "PREV_COUNT", "PREV_APPROVED_RATE", "PREV_REFUSED_RATE", "PREV_APP_CREDIT_RATIO_MEAN", "PREV_ANNUITY_MEAN", "PREV_CREDIT_MEAN", "PREV_DAYS_DECISION_MEAN", "PREV_DAYS_DECISION_MIN", "PREV_GOODS_PRICE_MEAN", "INST_PAYMENT_DIFF_MEAN", "INST_PAYMENT_DIFF_MAX", "INST_DAYS_ENTRY_DIFF_MEAN", "INST_LATE_PAYMENT_RATE", "INST_SHORT_PAYMENT_RATE", "INST_COUNT", "POS_MONTHS_COUNT", "POS_SK_DPD_MEAN", "POS_SK_DPD_MAX", "POS_DPD_RATE", "POS_CNT_INSTALMENT_MEAN", "CC_UTIL_RATE_MEAN", "CC_UTIL_RATE_MAX", "CC_DRAWING_RATE_MEAN", "CC_AMT_BALANCE_MEAN", "CC_COUNT", "CC_DPD_MEAN", "NLP_EMB_0", "NLP_EMB_1", "NLP_EMB_2", "NLP_EMB_3", "NLP_EMB_4", "NLP_EMB_5", "NLP_EMB_6", "NLP_EMB_7", "NLP_EMB_8", "NLP_EMB_9", "NLP_EMB_10", "NLP_EMB_11", "NLP_EMB_12", "NLP_EMB_13", "NLP_EMB_14", "NLP_EMB_15", "NLP_EMB_16", "NLP_EMB_17", "NLP_EMB_18", "NLP_EMB_19", "NLP_EMB_20", "NLP_EMB_21", "NLP_EMB_22", "NLP_EMB_23", "NLP_EMB_24", "NLP_EMB_25", "NLP_EMB_26", "NLP_EMB_27", "NLP_EMB_28", "NLP_EMB_29", "NLP_EMB_30", "NLP_EMB_31"]

kaggle_output/models/lgbm_fold_3.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0aa2a44a7ad2ae318c455318f0a14fd7cd88a74c3901eb15d259a7eebfd1acf
+size 10491639

kaggle_output/models/lgbm_fold_4.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fbfdb76aa33054e98456dd5578d28bbc349e85bf6e8ffc2d121d69044667ad2
+size 11224305

kaggle_output/models/lgbm_fold_5.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a8fcd6417da7b9ca0c124315a36dabb3e571151e27c40dd84b354ef58c81a65
+size 11612984

kaggle_output/models/pca.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:211b22e835c628cec4a9ba9603bda0ebd974f58ba0fb3631ff1d606570d0d75b
+size 52301

kaggle_output/models/scaler.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b776bd1834cd9402f94f6ca4285fd7dbcf81f0fdbe31b4195cdf0c2a106593a
+size 2199

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/files/config.yaml

@@ -0,0 +1,80 @@
+_wandb:
+    value:
+        cli_version: 0.25.0
+        e:
+            jyd6p8uwdd5m81jhhfteje7lwq5s32ua:
+                codePath: kaggle.ipynb
+                cpu_count: 2
+                cpu_count_logical: 4
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "8656922775552"
+                        used: "7347648929792"
+                email: suvraadeep@gmail.com
+                executable: /usr/bin/python3
+                gpu: Tesla T4
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Turing
+                      cudaCores: 2560
+                      memoryTotal: "16106127360"
+                      name: Tesla T4
+                      uuid: GPU-c7ebff42-1f98-33a6-9169-a1e3925c4f52
+                    - architecture: Turing
+                      cudaCores: 2560
+                      memoryTotal: "16106127360"
+                      name: Tesla T4
+                      uuid: GPU-4f057d35-8a41-ce55-5d7d-60c77ebb7156
+                host: 7a129c53b2b5
+                memory:
+                    total: "33662472192"
+                os: Linux-6.6.113+-x86_64-with-glibc2.35
+                program: kaggle.ipynb
+                python: CPython 3.12.12
+                root: /kaggle/working
+                startedAt: "2026-03-31T06:55:27.783331Z"
+                writerId: jyd6p8uwdd5m81jhhfteje7lwq5s32ua
+        m: []
+        python_version: 3.12.12
+        t:
+            "1":
+                - 1
+                - 5
+                - 6
+                - 8
+                - 11
+                - 35
+                - 49
+                - 53
+                - 54
+                - 71
+                - 75
+                - 105
+            "2":
+                - 1
+                - 5
+                - 6
+                - 8
+                - 11
+                - 35
+                - 49
+                - 53
+                - 54
+                - 71
+                - 75
+                - 105
+            "3":
+                - 2
+                - 13
+                - 15
+                - 16
+            "4": 3.12.12
+            "5": 0.25.0
+            "6": 4.41.2
+            "8":
+                - 1
+                - 2
+                - 12
+            "12": 0.25.0
+            "13": linux-x86_64

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/files/media/table/feature_importance_6_9280c5e00d174ed85360.table.json

@@ -0,0 +1 @@
+{"columns": ["feature", "importance"], "data": [["EXT_SOURCE_MEAN", 88957.16981611252], ["CREDIT_TERM", 11508.994362545014], ["GOODS_CREDIT_RATIO", 10540.86078901291], ["BUREAU_DEBT_CREDIT_RATIO_MAX", 10539.967162036895], ["EXT2_EXT3_INTERACTION", 8242.227200603485], ["POS_CNT_INSTALMENT_MEAN", 7895.31110868454], ["NLP_EMB_1", 6870.842185974121], ["EXT_SOURCE_MIN", 6497.98137922287], ["INST_LATE_PAYMENT_RATE", 6119.231726264954], ["POS_MONTHS_COUNT", 5955.207735443115], ["PREV_ANNUITY_MEAN", 5747.939696884156], ["AMT_ANNUITY", 5663.237604904175], ["EXT_SOURCE_3", 5611.383906459809], ["PREV_DAYS_DECISION_MIN", 5374.772545909882], ["EXT_SOURCE_STD", 5298.635174560547], ["BUREAU_DEBT_CREDIT_RATIO_MEAN", 5251.993024539947], ["PREV_REFUSED_RATE", 5157.323820114136], ["BUREAU_DAYS_CREDIT_MAX", 4958.93277130127], ["PREV_APP_CREDIT_RATIO_MEAN", 4918.679397964477], ["DAYS_ID_PUBLISH", 4889.688020515442]]}

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/files/output.log

@@ -0,0 +1,12 @@
+[200]	valid_0's auc: 0.775678
+  Fold 1 | AUC: 0.77593 | Best iter: 169
+[200]	valid_0's auc: 0.784365
+  Fold 2 | AUC: 0.78506 | Best iter: 266
+[200]	valid_0's auc: 0.778904
+  Fold 3 | AUC: 0.77896 | Best iter: 206
+[200]	valid_0's auc: 0.784897
+  Fold 4 | AUC: 0.78566 | Best iter: 297
+[200]	valid_0's auc: 0.773547
+  Fold 5 | AUC: 0.77408 | Best iter: 227
+
+🏆 OOF AUC: 0.77992 ± 0.00470

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/files/requirements.txt

@@ -0,0 +1,974 @@
+setuptools==75.2.0
+types-setuptools==80.10.0.20260124
+requirements-parser==0.9.0
+pip==24.1.2
+logistro==2.0.1
+tokenizers==0.19.1
+huggingface_hub==0.36.2
+scikit-learn==1.4.2
+xgboost==2.0.3
+sentence-transformers==2.7.0
+choreographer==1.2.1
+lightgbm==4.3.0
+pytest-timeout==2.4.0
+numpy==1.26.4
+category-encoders==2.6.3
+scipy==1.11.4
+river==0.23.0
+optuna-integration==4.8.0
+kaleido==1.2.0
+transformers==4.41.2
+plotly==6.6.0
+pytools==2025.2.5
+pycuda==2026.1
+siphash24==1.8
+protobuf==5.29.5
+torchtune==0.6.1
+learntools==0.3.5
+rouge_score==0.1.2
+pyclipper==1.4.0
+urwid_readline==0.15.1
+h2o==3.46.0.10
+rfc3161-client==1.0.5
+blake3==1.0.8
+mpld3==0.5.12
+qgrid==1.3.1
+ConfigSpace==1.2.2
+woodwork==0.31.0
+ujson==5.12.0
+y-py==0.6.2
+ipywidgets==8.1.5
+scikit-multilearn==0.2.0
+lightning-utilities==0.15.3
+pytesseract==0.3.13
+Cartopy==0.25.0
+odfpy==1.4.1
+Boruta==0.4.3
+docstring-to-markdown==0.17
+torchinfo==1.8.0
+clint==0.5.1
+comm==0.2.3
+Deprecated==1.3.1
+pymongo==4.16.0
+tensorflow-io-gcs-filesystem==0.37.1
+jmespath==1.1.0
+pygltflib==1.16.5
+keras-core==0.1.7
+pandas==2.3.3
+securesystemslib==1.3.1
+ghapi==1.0.11
+qtconsole==5.7.1
+pyemd==2.0.0
+pandas-profiling==3.6.6
+nilearn==0.13.1
+in-toto-attestation==0.9.3
+a2a-sdk==0.3.25
+keras-tuner==1.4.8
+fastuuid==0.14.0
+scikit-surprise==1.1.4
+vtk==9.3.1
+jupyter-ydoc==0.2.5
+aiofiles==22.1.0
+pytokens==0.4.1
+featuretools==1.31.0
+plotly-express==0.4.1
+marshmallow==3.26.2
+easyocr==1.7.2
+ppft==1.7.8
+openslide-bin==4.0.0.13
+fuzzywuzzy==0.18.0
+id==1.6.1
+openslide-python==1.4.3
+kaggle-environments==1.27.3
+pyarrow==23.0.1
+pandasql==0.7.3
+update-checker==0.18.0
+pathos==0.3.2
+jupyter_server_fileid==0.9.3
+fasttext==0.9.3
+coverage==7.13.5
+s3fs==2026.2.0
+stopit==1.1.2
+haversine==2.9.0
+jupyter_server==2.12.5
+geojson==3.2.0
+botocore==1.42.70
+fury==0.12.0
+ipympl==0.10.0
+ipython_pygments_lexers==1.1.1
+olefile==0.47
+jupyter_server_proxy==4.4.0
+datasets==4.8.3
+pytorch-ignite==0.5.3
+xvfbwrapper==0.2.22
+daal==2025.11.0
+open_spiel==1.6.12
+jupyter-lsp==1.5.1
+trx-python==0.4.0
+gpxpy==1.6.2
+papermill==2.7.0
+simpervisor==1.0.0
+kagglehub==1.0.0
+mlcrate==0.2.0
+kaggle==2.0.0
+dask-jobqueue==0.9.0
+model-signing==1.1.1
+jupyterlab==3.6.8
+args==0.1.0
+ImageHash==4.3.2
+typing-inspect==0.9.0
+PyUpSet==0.1.1.post7
+dacite==1.9.2
+pycryptodome==3.23.0
+google-cloud-videointelligence==2.18.0
+visions==0.8.1
+deap==1.4.3
+lml==0.2.0
+jiter==0.10.0
+ypy-websocket==0.8.4
+cytoolz==1.1.0
+path.py==12.5.0
+tensorflow-io==0.37.1
+wavio==0.0.9
+pdf2image==1.17.0
+line_profiler==5.0.2
+fsspec==2026.2.0
+aiobotocore==3.3.0
+optuna==4.8.0
+fastgit==0.0.4
+litellm==1.82.4
+pyLDAvis==3.4.1
+Janome==0.5.0
+langid==1.1.6
+sigstore-models==0.0.6
+pokerkit==0.6.3
+pyaml==26.2.1
+scikit-plot==0.3.7
+nbdev==3.0.12
+simpleitk==2.5.3
+ml_collections==1.1.0
+filetype==1.2.0
+Wand==0.7.0
+jupyter_server_ydoc==0.8.0
+pyjson5==2.0.0
+email-validator==2.3.0
+execnb==0.1.18
+colorama==0.4.6
+ruamel.yaml==0.19.1
+python-lsp-server==1.14.0
+black==26.3.1
+PyArabic==0.6.15
+gymnasium==1.2.0
+path==17.1.1
+gensim==4.4.0
+pypdf==6.9.1
+TPOT==1.1.0
+Pympler==1.1
+bayesian-optimization==3.2.1
+nbconvert==6.4.5
+kornia==0.8.2
+pathspec==1.0.4
+pybind11==3.0.2
+sigstore==4.2.0
+funcy==2.0
+func_timeout==4.3.5
+testpath==0.6.0
+aioitertools==0.13.0
+google-cloud-vision==3.12.1
+ray==2.54.0
+kornia_rs==0.1.10
+traitlets==5.14.3
+gymnax==0.0.8
+dnspython==2.8.0
+chex==0.1.90
+gym==0.26.2
+nbclient==0.5.13
+ydata-profiling==4.18.1
+POT==0.9.6.post1
+deepdiff==8.6.2
+squarify==0.4.4
+dataclasses-json==0.6.7
+pettingzoo==1.24.0
+pytorch-lightning==2.6.1
+segment_anything==1.0
+emoji==2.15.0
+python-bidi==0.6.7
+rgf-python==3.12.0
+ninja==1.13.0
+widgetsnbextension==4.0.15
+minify_html==0.18.1
+urwid==3.0.5
+jedi==0.19.2
+jupyterlab-lsp==3.10.2
+python-lsp-jsonrpc==1.1.2
+QtPy==2.4.3
+pydicom==3.0.1
+multimethod==1.12
+torchmetrics==1.9.0
+asttokens==3.0.1
+docker==7.1.0
+dask-expr==2.0.0
+s3transfer==0.16.0
+build==1.4.0
+Shimmy==2.0.0
+igraph==1.0.0
+puremagic==2.1.0
+jupyterlab_server==2.28.0
+isoweek==1.3.3
+texttable==1.7.0
+kt-legacy==1.0.5
+orderly-set==5.5.0
+pyexcel-io==0.6.7
+catboost==1.2.10
+kagglesdk==0.1.16
+mamba==0.11.3
+dipy==1.12.0
+colorlog==6.10.1
+asn1crypto==1.5.1
+pyexcel-ods==0.6.0
+lime==0.2.0.1
+pox==0.3.7
+rfc8785==0.1.4
+sigstore-rekor-types==0.0.18
+cesium==0.12.4
+boto3==1.42.70
+tuf==6.0.0
+hep_ml==0.8.0
+pyproject_hooks==1.2.0
+phik==0.12.5
+pudb==2025.1.5
+mne==1.11.0
+keras-cv==0.9.0
+dill==0.4.1
+gatspy==0.3
+scikit-learn-intelex==2025.11.0
+onnx==1.20.1
+scikit-optimize==0.10.2
+mypy_extensions==1.1.0
+mistune==0.8.4
+json5==0.13.0
+google-colab==1.0.0
+psutil==5.9.5
+jsonschema==4.26.0
+astunparse==1.6.3
+pycocotools==2.0.11
+lxml==6.0.2
+ipython==7.34.0
+oauthlib==3.3.1
+grpc-google-iam-v1==0.14.3
+array_record==0.8.3
+PuLP==3.3.0
+nvidia-cuda-runtime-cu12==12.8.90
+dask-cuda==26.2.0
+immutabledict==4.3.1
+peewee==4.0.0
+fiona==1.10.1
+aiosignal==1.4.0
+libclang==18.1.1
+annotated-types==0.7.0
+spreg==1.8.5
+grain==0.2.15
+geemap==0.35.3
+patsy==1.0.2
+imagesize==1.4.1
+py-cpuinfo==9.0.0
+pyzmq==26.2.1
+nvidia-cufile-cu12==1.13.1.3
+multidict==6.7.1
+srsly==2.5.2
+intel-openmp==2025.3.2
+uuid_utils==0.14.1
+google-cloud-language==2.19.0
+soxr==1.0.0
+jupyterlab_pygments==0.3.0
+backcall==0.2.0
+tensorflow-hub==0.16.1
+google==3.0.0
+requests-oauthlib==2.0.0
+dopamine_rl==4.1.2
+overrides==7.7.0
+db-dtypes==1.5.0
+jeepney==0.9.0
+langgraph-sdk==0.3.9
+ipython-genutils==0.2.0
+nvidia-cuda-cupti-cu12==12.8.90
+libcugraph-cu12==26.2.0
+catalogue==2.0.10
+beautifulsoup4==4.13.5
+nvidia-ml-py==13.590.48
+sphinxcontrib-devhelp==2.0.0
+partd==1.4.2
+sklearn-pandas==2.2.0
+sphinxcontrib-qthelp==2.0.0
+google-cloud-spanner==3.63.0
+h5py==3.15.1
+python-box==7.4.1
+distributed-ucxx-cu12==0.48.0
+xlrd==2.0.2
+branca==0.8.2
+chardet==5.2.0
+pycairo==1.29.0
+Authlib==1.6.8
+cuda-core==0.3.2
+sentencepiece==0.2.1
+nvidia-cusparselt-cu12==0.7.1
+matplotlib-venn==1.1.2
+scooby==0.11.0
+fqdn==1.5.1
+gin-config==0.5.0
+ipython-sql==0.5.0
+toml==0.10.2
+PyOpenGL==3.1.10
+weasel==0.4.3
+jsonpointer==3.0.0
+google-auth-httplib2==0.3.0
+spint==1.0.7
+nvtx==0.2.14
+websocket-client==1.9.0
+torchao==0.10.0
+splot==1.1.7
+langgraph-checkpoint==4.0.0
+alabaster==1.0.0
+jaxlib==0.7.2
+google-resumable-media==2.8.0
+namex==0.1.0
+quantecon==0.11.0
+nvidia-cuda-cccl-cu12==12.9.27
+google-cloud-aiplatform==1.138.0
+treelite==4.6.1
+google-cloud-resource-manager==1.16.0
+jupyter_core==5.9.1
+spacy-legacy==3.0.12
+librosa==0.11.0
+ibis-framework==9.5.0
+requests-toolbelt==1.0.0
+smart_open==7.5.1
+tensorflow-metadata==1.17.3
+pysal==25.7
+highspy==1.13.1
+click==8.3.1
+markdown-it-py==4.0.0
+nvidia-cusolver-cu12==11.7.3.90
+cupy-cuda12x==14.0.1
+imutils==0.5.4
+grpclib==0.4.9
+opt_einsum==3.4.0
+folium==0.20.0
+moviepy==1.0.3
+opencv-python==4.13.0.92
+en_core_web_sm==3.8.0
+tensorflow-text==2.19.0
+langchain-core==1.2.15
+yarl==1.22.0
+spacy==3.8.11
+importlib_resources==6.5.2
+peft==0.18.1
+lazy_loader==0.4
+polars-runtime-32==1.35.2
+pylibcudf-cu12==26.2.1
+bigquery-magics==0.10.3
+spanner-graph-notebook==1.1.8
+sqlglot==25.20.2
+linkify-it-py==2.0.3
+types-pytz==2025.2.0.20251108
+tifffile==2026.2.20
+tsfresh==0.21.1
+nbclassic==1.3.3
+scikit-image==0.25.2
+tensorflow_decision_forests==1.12.0
+simsimd==6.5.13
+isoduration==20.11.0
+momepy==0.11.0
+pytest==8.4.2
+nvidia-cuda-nvcc-cu12==12.5.82
+cuda-bindings==12.9.4
+torchsummary==1.5.1
+earthengine-api==1.5.24
+webencodings==0.5.1
+optree==0.19.0
+jax-cuda12-pjrt==0.7.2
+langchain==1.2.10
+safehttpx==0.1.7
+holidays==0.91
+google-cloud-firestore==2.23.0
+fastjsonschema==2.21.2
+pymc==5.28.0
+pydantic==2.12.3
+jaraco.context==6.1.0
+pyogrio==0.12.1
+numba-cuda==0.22.2
+fonttools==4.61.1
+httpimport==1.4.1
+rsa==4.9.1
+tomlkit==0.13.3
+entrypoints==0.4
+anyio==4.12.1
+charset-normalizer==3.4.4
+pooch==1.9.0
+libcuml-cu12==26.2.0
+astropy-iers-data==0.2026.2.23.0.48.33
+ipyleaflet==0.20.0
+cryptography==43.0.3
+missingno==0.5.2
+langgraph==1.0.9
+pandas-datareader==0.10.0
+pyviz_comms==3.0.6
+cycler==0.12.1
+tensorboard==2.19.0
+gast==0.7.0
+jax-cuda12-plugin==0.7.2
+platformdirs==4.9.2
+google-genai==1.64.0
+inflect==7.5.0
+httplib2==0.31.2
+h11==0.16.0
+alembic==1.18.4
+multitasking==0.0.12
+rmm-cu12==26.2.0
+cvxpy==1.6.7
+affine==2.4.0
+cuml-cu12==26.2.0
+pyparsing==3.3.2
+cffi==2.0.0
+h5netcdf==1.8.1
+Markdown==3.10.2
+google-cloud-translate==3.24.0
+rpy2==3.5.17
+regex==2025.11.3
+tf_keras==2.19.0
+google-auth==2.47.0
+nvidia-libnvcomp-cu12==5.1.0.21
+Send2Trash==2.1.0
+cymem==2.0.13
+pylibraft-cu12==26.2.0
+shap==0.50.0
+shapely==2.1.2
+psygnal==0.15.1
+uri-template==1.3.0
+parso==0.8.6
+webcolors==25.10.0
+nltk==3.9.1
+atpublic==5.1
+ImageIO==2.37.2
+sphinxcontrib-applehelp==2.0.0
+bigframes==2.35.0
+pydot==4.0.1
+onemkl-license==2025.3.1
+treescope==0.1.10
+tcmlib==1.4.1
+opentelemetry-sdk==1.38.0
+tiktoken==0.12.0
+nibabel==5.3.3
+multiprocess==0.70.16
+typing_extensions==4.15.0
+PyYAML==6.0.3
+defusedxml==0.7.1
+sphinxcontrib-serializinghtml==2.0.0
+bleach==6.3.0
+tenacity==9.1.4
+python-utils==3.9.1
+google-cloud-bigquery==3.40.1
+google-cloud-bigquery-connection==1.20.0
+opentelemetry-resourcedetector-gcp==1.11.0a0
+ormsgpack==1.12.2
+pydotplus==2.0.2
+pycryptodomex==3.23.0
+openai==2.23.0
+matplotlib==3.10.0
+ml_dtypes==0.5.4
+uvloop==0.22.1
+google-pasta==0.2.0
+giddy==2.3.8
+ipyparallel==8.8.0
+keras==3.10.0
+cuvs-cu12==26.2.0
+mcp==1.26.0
+spacy-loggers==1.0.5
+google-cloud-logging==3.13.0
+rfc3987-syntax==1.1.0
+google-ai-generativelanguage==0.6.15
+keras-hub==0.21.1
+pydata-google-auth==1.9.1
+absl-py==1.4.0
+ydf==0.15.0
+narwhals==2.17.0
+nvidia-cusparse-cu12==12.5.8.93
+openpyxl==3.1.5
+nvidia-cublas-cu12==12.8.4.1
+roman-numerals==4.1.0
+vega-datasets==0.9.0
+mpmath==1.3.0
+etils==1.13.0
+osqp==1.1.1
+traittypes==0.2.3
+opentelemetry-exporter-gcp-monitoring==1.11.0a0
+graphviz==0.21
+google-cloud-trace==1.18.0
+einops==0.8.2
+torchdata==0.11.0
+jax==0.7.2
+cachetools==6.2.6
+aiohappyeyeballs==2.6.1
+annotated-doc==0.0.4
+starlette==0.52.1
+fastapi==0.133.0
+typer==0.24.1
+duckdb==1.3.2
+blinker==1.9.0
+referencing==0.37.0
+googledrivedownloader==1.1.0
+GDAL==3.8.4
+cuda-python==12.9.4
+pycparser==3.0
+et_xmlfile==2.0.0
+jieba==0.42.1
+zict==3.0.0
+hyperopt==0.2.7
+python-louvain==0.16
+SQLAlchemy==2.0.47
+cuda-toolkit==12.8.1
+PyDrive2==1.21.3
+roman-numerals-py==4.1.0
+urllib3==2.5.0
+jaraco.functools==4.4.0
+optax==0.2.7
+pyOpenSSL==24.2.1
+jupyter-console==6.6.3
+libkvikio-cu12==26.2.0
+gspread==6.2.1
+docstring_parser==0.17.0
+albumentations==2.0.8
+jupytext==1.19.1
+seaborn==0.13.2
+librmm-cu12==26.2.0
+cons==0.4.7
+matplotlib-inline==0.2.1
+pynndescent==0.6.0
+stringzilla==4.6.0
+flatbuffers==25.12.19
+omegaconf==2.3.0
+umap-learn==0.5.11
+progressbar2==4.5.0
+pexpect==4.9.0
+torchcodec==0.10.0+cu128
+ptyprocess==0.7.0
+pygame==2.6.1
+kiwisolver==1.4.9
+Cython==3.0.12
+shellingham==1.5.4
+soupsieve==2.8.3
+snowballstemmer==3.0.1
+propcache==0.4.1
+ucxx-cu12==0.48.0
+nbformat==5.10.4
+python-snappy==0.7.3
+rasterstats==0.20.0
+bqplot==0.12.45
+nest-asyncio==1.6.0
+opencv-python-headless==4.13.0.92
+notebook==6.5.7
+flax==0.11.2
+google-cloud-functions==1.22.0
+multipledispatch==1.0.0
+googleapis-common-protos==1.72.0
+eerepr==0.1.2
+torchaudio==2.10.0+cu128
+locket==1.0.0
+prettytable==3.17.0
+pygit2==1.19.1
+fastai==2.8.7
+msgpack==1.1.2
+clarabel==0.11.1
+cligj==0.7.2
+google-cloud-secret-manager==2.26.0
+spglm==1.1.0
+ipytree==0.2.2
+termcolor==3.3.0
+tweepy==4.16.0
+google-cloud-core==2.5.0
+dataproc-spark-connect==1.0.2
+mkl==2025.3.1
+umf==1.0.3
+textblob==0.19.0
+firebase-admin==6.9.0
+simple-parsing==0.1.8
+debugpy==1.8.15
+google-cloud-discoveryengine==0.13.12
+fastcore==1.12.16
+decorator==4.4.2
+pickleshare==0.7.5
+rasterio==1.5.0
+networkx==3.6.1
+typer-slim==0.24.0
+wasabi==1.1.3
+mgwr==2.2.1
+hdbscan==0.8.41
+pydub==0.25.1
+tobler==0.13.0
+more-itertools==10.8.0
+keyrings.google-artifactregistry-auth==1.1.2
+cloudpickle==3.1.2
+nvidia-nvtx-cu12==12.8.90
+fastlite==0.2.4
+colorcet==3.1.0
+lark==1.3.1
+antlr4-python3-runtime==4.9.3
+keras-nlp==0.21.1
+music21==9.9.1
+Pygments==2.19.2
+triton==3.6.0
+toolz==0.12.1
+python-slugify==8.0.4
+sqlparse==0.5.5
+jupyter-leaflet==0.20.0
+gym-notices==0.1.0
+torchvision==0.25.0+cu128
+prophet==1.3.0
+google-cloud-datastore==2.23.0
+semantic-version==2.10.0
+fastprogress==1.1.5
+etuples==0.3.10
+pyspark==4.0.2
+orjson==3.11.7
+terminado==0.18.1
+accelerate==1.12.0
+panel==1.8.7
+apswutils==0.1.2
+pyproj==3.7.2
+sphinxcontrib-htmlhelp==2.1.0
+certifi==2026.1.4
+grpc-interceptor==0.15.4
+pyasn1==0.6.2
+geocoder==1.38.1
+idna==3.11
+mizani==0.13.5
+jupyter_server_terminals==0.5.4
+httpcore==1.0.9
+pyasn1_modules==0.4.2
+ffmpy==1.0.0
+pyperclip==1.11.0
+safetensors==0.7.0
+ndindex==1.10.1
+tblib==3.2.2
+docutils==0.21.2
+scs==3.2.11
+distro==1.9.0
+tf-slim==1.1.0
+babel==2.18.0
+google-cloud-pubsub==2.35.0
+google-api-python-client==2.190.0
+tzlocal==5.3.1
+groovy==0.1.2
+plum-dispatch==2.7.1
+dask==2026.1.1
+blosc2==4.0.0
+sqlalchemy-spanner==1.17.2
+orbax-checkpoint==0.11.33
+wandb==0.25.0
+geopandas==1.1.2
+proglog==0.1.12
+python-dateutil==2.9.0.post0
+tzdata==2025.3
+editdistance==0.8.1
+langsmith==0.7.6
+xarray-einstats==0.10.0
+pydantic_core==2.41.4
+tabulate==0.9.0
+mmh3==5.2.0
+sentry-sdk==2.53.0
+spopt==0.7.0
+dlib==19.24.6
+community==1.0.0b1
+tensorflow==2.19.0
+ale-py==0.11.2
+murmurhash==1.0.15
+notebook_shim==0.2.4
+mdurl==0.1.2
+diffusers==0.36.0
+requests==2.32.4
+Flask==3.1.3
+prometheus_client==0.24.1
+uvicorn==0.41.0
+logical-unification==0.4.7
+soundfile==0.13.1
+itsdangerous==2.2.0
+jsonpatch==1.33
+plotnine==0.14.5
+distributed==2026.1.1
+google-auth-oauthlib==1.2.4
+gdown==5.2.1
+brotli==1.2.0
+py4j==0.10.9.9
+pytensor==2.38.0
+text-unidecode==1.3
+yfinance==0.2.66
+arviz==0.22.0
+cudf-cu12==26.2.1
+wordcloud==1.9.6
+jaraco.classes==3.4.0
+albucore==0.0.24
+python-dotenv==1.2.1
+uritemplate==4.2.0
+nx-cugraph-cu12==26.2.0
+raft-dask-cu12==26.2.0
+hpack==4.1.0
+numexpr==2.14.1
+pydantic-settings==2.13.1
+rapids-logger==0.2.3
+cmake==3.31.10
+pillow==11.3.0
+jsonschema-specifications==2025.9.1
+tables==3.10.2
+google-cloud-storage==3.9.0
+mapclassify==2.10.0
+altair==5.5.0
+filelock==3.24.3
+google-cloud-appengine-logging==1.8.0
+cufflinks==0.17.3
+cvxopt==1.3.2
+six==1.17.0
+watchdog==6.0.0
+sse-starlette==3.2.0
+PySocks==1.7.1
+jupyterlab_widgets==3.0.16
+spaghetti==1.7.6
+intel-cmplr-lib-ur==2025.3.2
+uc-micro-py==1.0.3
+Sphinx==8.2.3
+PyJWT==2.11.0
+google-cloud-bigtable==2.35.0
+numba==0.60.0
+httptools==0.7.1
+rich==13.9.4
+pointpats==2.5.5
+watchfiles==1.1.1
+promise==2.3
+polars==1.35.2
+greenlet==3.3.2
+rfc3986-validator==0.1.1
+threadpoolctl==3.6.0
+opentelemetry-exporter-otlp-proto-http==1.38.0
+libcuvs-cu12==26.2.0
+sniffio==1.3.1
+pylibcugraph-cu12==26.2.0
+holoviews==1.22.1
+pandas-gbq==0.30.0
+frozenlist==1.8.0
+google-crc32c==1.8.0
+torch==2.10.0+cu128
+ipyevents==2.0.4
+libucxx-cu12==0.48.0
+cramjam==2.11.0
+opentelemetry-exporter-otlp-proto-common==1.38.0
+wurlitzer==3.1.1
+confection==0.1.5
+stanio==0.5.1
+easydict==1.13
+argon2-cffi==25.1.0
+llvmlite==0.43.0
+humanize==4.15.0
+rapids-dask-dependency==26.2.0
+argon2-cffi-bindings==25.1.0
+future==1.0.0
+rpds-py==0.30.0
+psycopg2==2.9.11
+iniconfig==2.3.0
+jupyter-events==0.12.0
+nvidia-nccl-cu12==2.27.5
+GitPython==3.1.46
+joblib==1.5.3
+beartype==0.22.9
+hf-xet==1.3.0
+Bottleneck==1.4.2
+apsw==3.51.2.0
+bokeh==3.8.2
+google-cloud-dataproc==5.25.0
+nvidia-cuda-nvrtc-cu12==12.8.93
+colour==0.1.5
+zipp==3.23.0
+blis==1.3.3
+click-plugins==1.1.1.2
+httpx-sse==0.4.3
+nvidia-nvshmem-cu12==3.4.5
+sphinxcontrib-jsmath==1.0.1
+prompt_toolkit==3.0.52
+esda==2.8.1
+param==2.3.2
+google-cloud-speech==2.36.1
+portpicker==1.5.2
+PyWavelets==1.9.0
+google-cloud-monitoring==2.29.1
+Farama-Notifications==0.0.4
+pytz==2025.2
+MarkupSafe==3.0.3
+pyomo==6.10.0
+packaging==26.0
+betterproto==2.0.0b6
+libraft-cu12==26.2.0
+typeguard==4.5.1
+imbalanced-learn==0.14.1
+google-adk==1.25.1
+CacheControl==0.14.4
+ipykernel==6.17.1
+jsonpickle==4.1.1
+xyzservices==2025.11.0
+websockets==15.0.1
+PyGObject==3.48.2
+pandas-stubs==2.2.2.240909
+proto-plus==1.27.1
+segregation==2.5.3
+ratelim==0.1.6
+miniKanren==1.0.5
+geographiclib==2.1
+Jinja2==3.1.6
+frozendict==2.4.7
+libcudf-cu12==26.2.1
+nvidia-cufft-cu12==11.3.3.83
+typing-inspection==0.4.2
+gradio_client==1.14.0
+simplejson==3.20.2
+ruff==0.15.2
+imageio-ffmpeg==0.6.0
+python-json-logger==4.0.0
+cucim-cu12==26.2.0
+jupyter_kernel_gateway==2.5.2
+contourpy==1.3.3
+google-api-core==2.30.0
+opencv-contrib-python==4.13.0.92
+nvidia-cudnn-cu12==9.10.2.21
+opentelemetry-proto==1.38.0
+dask-cudf-cu12==26.2.1
+nvidia-nvimgcodec-cu12==0.7.0.11
+statsmodels==0.14.6
+opentelemetry-exporter-gcp-trace==1.11.0
+deprecation==2.1.0
+tinycss2==1.4.0
+mdit-py-plugins==0.5.0
+tensorflow-datasets==4.9.9
+opentelemetry-api==1.38.0
+langgraph-prebuilt==1.0.8
+keyring==25.7.0
+inequality==1.1.2
+cyipopt==1.5.0
+sympy==1.14.0
+oauth2client==4.1.3
+python-fasthtml==0.12.47
+gspread-dataframe==4.0.0
+wcwidth==0.6.0
+geopy==2.4.1
+natsort==8.4.0
+timm==1.0.25
+rfc3339-validator==0.1.4
+stumpy==1.13.0
+parsy==2.2
+libucx-cu12==1.19.0
+pyerfa==2.0.1.5
+astropy==7.2.0
+curl_cffi==0.14.0
+xarray==2025.12.0
+preshed==3.0.12
+Werkzeug==3.1.6
+SecretStorage==3.5.0
+grpcio==1.78.1
+slicer==0.0.8
+cudf-polars-cu12==26.2.1
+aiosqlite==0.22.1
+grpcio-status==1.71.2
+libpysal==4.14.1
+gitdb==4.0.12
+hyperframe==6.1.0
+opentelemetry-semantic-conventions==0.59b0
+wheel==0.46.3
+h2==4.3.0
+google-cloud-audit-log==0.4.0
+tqdm==4.67.3
+httpx==0.28.1
+cloudpathlib==0.23.0
+thinc==8.3.10
+audioread==3.1.0
+fastdownload==0.0.7
+gcsfs==2025.3.0
+nvidia-nvjitlink-cu12==12.8.93
+access==1.1.10.post3
+tornado==6.5.1
+pandocfilters==1.5.1
+fasttransform==0.0.2
+nvidia-curand-cu12==10.3.9.90
+python-multipart==0.0.22
+yellowbrick==1.5
+jupyter_client==7.4.9
+google-generativeai==0.8.6
+blobfile==3.2.0
+importlib_metadata==8.7.1
+tensorboard-data-server==0.7.2
+attrs==25.4.0
+tbb==2022.3.1
+pluggy==1.6.0
+cuda-pathfinder==1.3.5
+rtree==1.4.1
+arrow==1.4.0
+wrapt==2.1.1
+anywidget==0.9.21
+mlxtend==0.23.4
+smmap==5.0.2
+aiohttp==3.13.3
+opentelemetry-exporter-gcp-logging==1.11.0a0
+sortedcontainers==2.4.0
+pyshp==3.0.3
+sklearn-compat==0.1.5
+xxhash==3.6.0
+zstandard==0.25.0
+Mako==1.3.10
+google-cloud-iam==2.21.0
+autograd==1.8.0
+glob2==0.7
+tensorstore==0.1.81
+tensorflow-probability==0.25.0
+colorlover==0.3.0
+ipyfilechooser==0.6.0
+gradio==5.50.0
+cmdstanpy==1.3.0
+dm-tree==0.1.9
+html5lib==1.1
+python-apt==0.0.0
+PyGObject==3.42.1
+blinker==1.4
+jeepney==0.7.1
+six==1.16.0
+oauthlib==3.2.0
+wadllib==1.3.6
+launchpadlib==1.10.16
+dbus-python==1.2.18
+PyJWT==2.3.0
+importlib-metadata==4.6.4
+httplib2==0.20.2
+zipp==1.0.0
+pyparsing==2.4.7
+lazr.restfulclient==0.14.4
+SecretStorage==3.3.1
+distro==1.7.0
+lazr.uri==1.0.6
+more-itertools==8.10.0
+python-apt==2.4.0+ubuntu4.1
+cryptography==3.4.8
+keyring==23.5.0
+Markdown==3.3.6
+Mako==1.1.3
+MarkupSafe==2.0.1
+packaging==24.1
+inflect==7.3.1
+autocommand==2.2.2
+typeguard==4.3.0
+jaraco.text==3.12.1
+importlib_resources==6.4.0
+wheel==0.43.0
+zipp==3.19.2
+platformdirs==4.2.2
+importlib_metadata==8.0.0
+tomli==2.0.1
+jaraco.collections==5.1.0
+more-itertools==10.3.0
+typing_extensions==4.12.2
+backports.tarfile==1.2.0
+jaraco.functools==4.0.1
+jaraco.context==5.3.0

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/files/wandb-metadata.json

@@ -0,0 +1,42 @@
+{
+  "os":  "Linux-6.6.113+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.12.12",
+  "startedAt":  "2026-03-31T06:55:27.783331Z",
+  "program":  "kaggle.ipynb",
+  "codePath":  "kaggle.ipynb",
+  "email":  "suvraadeep@gmail.com",
+  "root":  "/kaggle/working",
+  "host":  "7a129c53b2b5",
+  "executable":  "/usr/bin/python3",
+  "cpu_count":  2,
+  "cpu_count_logical":  4,
+  "gpu":  "Tesla T4",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "8656922775552",
+      "used":  "7347648929792"
+    }
+  },
+  "memory":  {
+    "total":  "33662472192"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "Tesla T4",
+      "memoryTotal":  "16106127360",
+      "cudaCores":  2560,
+      "architecture":  "Turing",
+      "uuid":  "GPU-c7ebff42-1f98-33a6-9169-a1e3925c4f52"
+    },
+    {
+      "name":  "Tesla T4",
+      "memoryTotal":  "16106127360",
+      "cudaCores":  2560,
+      "architecture":  "Turing",
+      "uuid":  "GPU-4f057d35-8a41-ce55-5d7d-60c77ebb7156"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "jyd6p8uwdd5m81jhhfteje7lwq5s32ua"
+}

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/files/wandb-summary.json

@@ -0,0 +1 @@
+{"n_features":234,"_step":6,"fold_4_auc":0.785662839195165,"n_train":307511,"fold_1_auc":0.7759344239846245,"fold_5_auc":0.7740783393085308,"_runtime":340,"fold_std":0.004696875321178463,"feature_importance":{"log_mode":"IMMUTABLE","sha256":"9280c5e00d174ed85360bfef885ef1bdd68e2abd997bb9ce86ab5bad73c62e80","ncols":2,"nrows":20,"_type":"table-file","size":909,"artifact_path":"wandb-client-artifact://q52w3a226nb82x86h1fbmlu6pk4a1pfhhgo1ypln5p6f3m1jo0y139ufnfpw64fhzjscu59hx5ez5aidin3iwxikjs3mxvvucbscl5amw5sab6jnh1njmqdzb1qyswkv/feature_importance.table.json","_latest_artifact_path":"wandb-client-artifact://0qa79omgcechd1vsci896g88gfa40010jtogetx60ek4susdjv3aqe995rhq8ol14qba5im486ruer71vxou9p08yyebvgx3y8lwysph1yelxvlnmy0safqn6dtjvl31:latest/feature_importance.table.json","path":"media/table/feature_importance_6_9280c5e00d174ed85360.table.json"},"oof_auc":0.7799152308164182,"fold_3_auc":0.7789614162608197,"fold":5,"fold_2_auc":0.7850577118521481,"_wandb":{"runtime":340},"_timestamp":1.77494047535925e+09}

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/logs/debug-internal.log

@@ -0,0 +1,12 @@
+{"time":"2026-03-31T06:55:28.579493541Z","level":"INFO","msg":"stream: starting","core version":"0.25.0"}
+{"time":"2026-03-31T06:55:28.73430316Z","level":"INFO","msg":"stream: created new stream","id":"jxzjz5r3"}
+{"time":"2026-03-31T06:55:28.735368816Z","level":"INFO","msg":"handler: started","stream_id":"jxzjz5r3"}
+{"time":"2026-03-31T06:55:28.735514769Z","level":"INFO","msg":"stream: started","id":"jxzjz5r3"}
+{"time":"2026-03-31T06:55:28.735571676Z","level":"INFO","msg":"sender: started","stream_id":"jxzjz5r3"}
+{"time":"2026-03-31T06:55:28.735597161Z","level":"INFO","msg":"writer: started","stream_id":"jxzjz5r3"}
+{"time":"2026-03-31T07:01:16.531006817Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-03-31T07:01:16.691833668Z","level":"INFO","msg":"handler: operation stats","stats":{}}
+{"time":"2026-03-31T07:01:16.698883138Z","level":"INFO","msg":"stream: closing","id":"jxzjz5r3"}
+{"time":"2026-03-31T07:01:16.698899918Z","level":"INFO","msg":"handler: closed","stream_id":"jxzjz5r3"}
+{"time":"2026-03-31T07:01:16.698949589Z","level":"INFO","msg":"sender: closed","stream_id":"jxzjz5r3"}
+{"time":"2026-03-31T07:01:16.698960323Z","level":"INFO","msg":"stream: closed","id":"jxzjz5r3"}

kaggle_output/wandb/run-20260331_065527-jxzjz5r3/logs/debug.log

@@ -0,0 +1,29 @@
+2026-03-31 06:55:27,790 INFO    MainThread:177 [wandb_setup.py:_flush():81] Current SDK version is 0.25.0
+2026-03-31 06:55:27,790 INFO    MainThread:177 [wandb_setup.py:_flush():81] Configure stats pid to 177
+2026-03-31 06:55:27,790 INFO    MainThread:177 [wandb_setup.py:_flush():81] Loading settings from environment variables
+2026-03-31 06:55:27,790 INFO    MainThread:177 [wandb_init.py:setup_run_log_directory():717] Logging user logs to /kaggle/working/wandb/run-20260331_065527-jxzjz5r3/logs/debug.log
+2026-03-31 06:55:27,790 INFO    MainThread:177 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to /kaggle/working/wandb/run-20260331_065527-jxzjz5r3/logs/debug-internal.log
+2026-03-31 06:55:27,790 INFO    MainThread:177 [wandb_init.py:monkeypatch_ipython():636] configuring jupyter hooks <wandb.sdk.wandb_init._WandbInit object at 0x7e0b144d1940>
+2026-03-31 06:55:27,791 INFO    MainThread:177 [wandb_init.py:init():844] calling init triggers
+2026-03-31 06:55:27,791 INFO    MainThread:177 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
+config: {'_wandb': {}}
+2026-03-31 06:55:27,791 INFO    MainThread:177 [wandb_init.py:init():892] starting backend
+2026-03-31 06:55:28,558 INFO    MainThread:177 [wandb_init.py:init():895] sending inform_init request
+2026-03-31 06:55:28,569 INFO    MainThread:177 [wandb_init.py:init():903] backend started and connected
+2026-03-31 06:55:28,577 INFO    MainThread:177 [wandb_run.py:_label_probe_notebook():1333] probe notebook
+2026-03-31 06:55:34,034 INFO    MainThread:177 [wandb_init.py:init():973] updated telemetry
+2026-03-31 06:55:34,270 INFO    MainThread:177 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
+2026-03-31 06:55:34,618 INFO    MainThread:177 [wandb_init.py:init():1042] starting run threads in backend
+2026-03-31 06:55:35,321 INFO    MainThread:177 [wandb_run.py:_console_start():2524] atexit reg
+2026-03-31 06:55:35,321 INFO    MainThread:177 [wandb_run.py:_redirect():2373] redirect: wrap_raw
+2026-03-31 06:55:35,321 INFO    MainThread:177 [wandb_run.py:_redirect():2442] Wrapping output streams.
+2026-03-31 06:55:35,321 INFO    MainThread:177 [wandb_run.py:_redirect():2465] Redirects installed.
+2026-03-31 06:55:35,337 INFO    MainThread:177 [wandb_init.py:init():1082] run started, returning control to user process
+2026-03-31 07:01:15,359 INFO    MainThread:177 [wandb_run.py:_finish():2291] finishing run suvradeep-iit-guwahati-/credit-invisibility/jxzjz5r3
+2026-03-31 07:01:15,360 INFO    MainThread:177 [jupyter.py:save_history():435] not saving jupyter history
+2026-03-31 07:01:15,360 INFO    MainThread:177 [jupyter.py:save_ipynb():362] not saving jupyter notebook
+2026-03-31 07:01:15,360 INFO    MainThread:177 [wandb_init.py:_jupyter_teardown():621] cleaning up jupyter logic
+2026-03-31 07:01:15,360 INFO    MainThread:177 [wandb_run.py:_atexit_cleanup():2490] got exitcode: 0
+2026-03-31 07:01:15,361 INFO    MainThread:177 [wandb_run.py:_restore():2472] restore
+2026-03-31 07:01:15,361 INFO    MainThread:177 [wandb_run.py:_restore():2478] restore done
+2026-03-31 07:01:16,696 INFO    MainThread:177 [wandb_run.py:_footer_sync_info():3868] logging synced files

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/files/config.yaml

@@ -0,0 +1,122 @@
+_wandb:
+    value:
+        cli_version: 0.25.0
+        e:
+            s21p4sab0yaskqym8r37eeuqwjpzso9d:
+                codePath: kaggle.ipynb
+                cpu_count: 2
+                cpu_count_logical: 4
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "8656922775552"
+                        used: "7347653709824"
+                email: suvraadeep@gmail.com
+                executable: /usr/bin/python3
+                gpu: Tesla T4
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Turing
+                      cudaCores: 2560
+                      memoryTotal: "16106127360"
+                      name: Tesla T4
+                      uuid: GPU-c7ebff42-1f98-33a6-9169-a1e3925c4f52
+                    - architecture: Turing
+                      cudaCores: 2560
+                      memoryTotal: "16106127360"
+                      name: Tesla T4
+                      uuid: GPU-4f057d35-8a41-ce55-5d7d-60c77ebb7156
+                host: 7a129c53b2b5
+                memory:
+                    total: "33662472192"
+                os: Linux-6.6.113+-x86_64-with-glibc2.35
+                program: kaggle.ipynb
+                python: CPython 3.12.12
+                root: /kaggle/working
+                startedAt: "2026-03-31T09:47:28.010618Z"
+                writerId: s21p4sab0yaskqym8r37eeuqwjpzso9d
+        m: []
+        python_version: 3.12.12
+        t:
+            "1":
+                - 1
+                - 5
+                - 6
+                - 8
+                - 11
+                - 35
+                - 49
+                - 53
+                - 54
+                - 71
+                - 75
+                - 105
+            "2":
+                - 1
+                - 5
+                - 6
+                - 8
+                - 11
+                - 35
+                - 49
+                - 53
+                - 54
+                - 71
+                - 75
+                - 105
+            "3":
+                - 2
+                - 13
+                - 15
+                - 16
+            "4": 3.12.12
+            "5": 0.25.0
+            "6": 4.41.2
+            "8":
+                - 1
+                - 2
+                - 12
+            "12": 0.25.0
+            "13": linux-x86_64
+lgbm:
+    value:
+        bagging_fraction: 0.8170784332632994
+        bagging_freq: 1
+        boosting_type: gbdt
+        device: gpu
+        feature_fraction: 0.7123738038749523
+        force_col_wise: true
+        gpu_use_dp: false
+        learning_rate: 0.02273805573563183
+        max_bin: 255
+        max_depth: 9
+        metric: auc
+        min_child_samples: 82
+        min_gain_to_split: 0.0993578407670862
+        n_jobs: -1
+        num_leaves: 112
+        objective: binary
+        reg_alpha: 0.0002359137306347715
+        reg_lambda: 8.598737339212267
+        seed: 42
+        verbosity: -1
+n_folds:
+    value: 5
+xgb:
+    value:
+        colsample_bytree: 0.8226259232371986
+        device: cuda
+        eval_metric: auc
+        gamma: 0.26516420723672285
+        learning_rate: 0.028084447839949865
+        max_bin: 256
+        max_depth: 5
+        min_child_weight: 4
+        n_jobs: -1
+        objective: binary:logistic
+        reg_alpha: 8.576808828106026
+        reg_lambda: 6.598703417106125
+        scale_pos_weight: 1.3690073364184308
+        seed: 42
+        subsample: 0.6252365594243677
+        tree_method: hist

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/files/output.log

@@ -0,0 +1,65 @@
+
+🚀 Fold 1
+[100]	valid_0's auc: 0.765307
+[200]	valid_0's auc: 0.774084
+[300]	valid_0's auc: 0.777392
+[400]	valid_0's auc: 0.77907
+[500]	valid_0's auc: 0.780086
+[600]	valid_0's auc: 0.780793
+[700]	valid_0's auc: 0.781234
+[800]	valid_0's auc: 0.781458
+[900]	valid_0's auc: 0.781597
+  LGBM: 0.78174 | XGB: 0.78181
+
+🚀 Fold 2
+[100]	valid_0's auc: 0.773239
+[200]	valid_0's auc: 0.783192
+[300]	valid_0's auc: 0.786755
+[400]	valid_0's auc: 0.788592
+[500]	valid_0's auc: 0.78945
+[600]	valid_0's auc: 0.789662
+[700]	valid_0's auc: 0.789836
+[800]	valid_0's auc: 0.789936
+  LGBM: 0.79000 | XGB: 0.79209
+
+🚀 Fold 3
+[100]	valid_0's auc: 0.76562
+[200]	valid_0's auc: 0.774939
+[300]	valid_0's auc: 0.779674
+[400]	valid_0's auc: 0.7817
+[500]	valid_0's auc: 0.782736
+[600]	valid_0's auc: 0.78322
+[700]	valid_0's auc: 0.783531
+[800]	valid_0's auc: 0.783716
+[900]	valid_0's auc: 0.78377
+  LGBM: 0.78383 | XGB: 0.78411
+
+🚀 Fold 4
+[100]	valid_0's auc: 0.773374
+[200]	valid_0's auc: 0.782137
+[300]	valid_0's auc: 0.786103
+[400]	valid_0's auc: 0.78795
+[500]	valid_0's auc: 0.788917
+[600]	valid_0's auc: 0.789669
+[700]	valid_0's auc: 0.789976
+[800]	valid_0's auc: 0.790266
+[900]	valid_0's auc: 0.790474
+[1000]	valid_0's auc: 0.790344
+  LGBM: 0.79050 | XGB: 0.79068
+
+🚀 Fold 5
+[100]	valid_0's auc: 0.763112
+[200]	valid_0's auc: 0.773904
+[300]	valid_0's auc: 0.778297
+[400]	valid_0's auc: 0.780052
+[500]	valid_0's auc: 0.781518
+[600]	valid_0's auc: 0.782167
+[700]	valid_0's auc: 0.782807
+[800]	valid_0's auc: 0.782822
+[900]	valid_0's auc: 0.783009
+[1000]	valid_0's auc: 0.782948
+  LGBM: 0.78309 | XGB: 0.78362
+
+🔍 Optimizing blend weights...
+✅ Best weight → LGBM: 0.49, XGB: 0.51
+🏆 Best OOF AUC: 0.78731

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/files/requirements.txt

@@ -0,0 +1,974 @@
+setuptools==75.2.0
+types-setuptools==80.10.0.20260124
+requirements-parser==0.9.0
+pip==24.1.2
+logistro==2.0.1
+tokenizers==0.19.1
+huggingface_hub==0.36.2
+scikit-learn==1.4.2
+xgboost==2.0.3
+sentence-transformers==2.7.0
+choreographer==1.2.1
+lightgbm==4.3.0
+pytest-timeout==2.4.0
+numpy==1.26.4
+category-encoders==2.6.3
+scipy==1.11.4
+river==0.23.0
+optuna-integration==4.8.0
+kaleido==1.2.0
+transformers==4.41.2
+plotly==6.6.0
+pytools==2025.2.5
+pycuda==2026.1
+siphash24==1.8
+protobuf==5.29.5
+torchtune==0.6.1
+learntools==0.3.5
+rouge_score==0.1.2
+pyclipper==1.4.0
+urwid_readline==0.15.1
+h2o==3.46.0.10
+rfc3161-client==1.0.5
+blake3==1.0.8
+mpld3==0.5.12
+qgrid==1.3.1
+ConfigSpace==1.2.2
+woodwork==0.31.0
+ujson==5.12.0
+y-py==0.6.2
+ipywidgets==8.1.5
+scikit-multilearn==0.2.0
+lightning-utilities==0.15.3
+pytesseract==0.3.13
+Cartopy==0.25.0
+odfpy==1.4.1
+Boruta==0.4.3
+docstring-to-markdown==0.17
+torchinfo==1.8.0
+clint==0.5.1
+comm==0.2.3
+Deprecated==1.3.1
+pymongo==4.16.0
+tensorflow-io-gcs-filesystem==0.37.1
+jmespath==1.1.0
+pygltflib==1.16.5
+keras-core==0.1.7
+pandas==2.3.3
+securesystemslib==1.3.1
+ghapi==1.0.11
+qtconsole==5.7.1
+pyemd==2.0.0
+pandas-profiling==3.6.6
+nilearn==0.13.1
+in-toto-attestation==0.9.3
+a2a-sdk==0.3.25
+keras-tuner==1.4.8
+fastuuid==0.14.0
+scikit-surprise==1.1.4
+vtk==9.3.1
+jupyter-ydoc==0.2.5
+aiofiles==22.1.0
+pytokens==0.4.1
+featuretools==1.31.0
+plotly-express==0.4.1
+marshmallow==3.26.2
+easyocr==1.7.2
+ppft==1.7.8
+openslide-bin==4.0.0.13
+fuzzywuzzy==0.18.0
+id==1.6.1
+openslide-python==1.4.3
+kaggle-environments==1.27.3
+pyarrow==23.0.1
+pandasql==0.7.3
+update-checker==0.18.0
+pathos==0.3.2
+jupyter_server_fileid==0.9.3
+fasttext==0.9.3
+coverage==7.13.5
+s3fs==2026.2.0
+stopit==1.1.2
+haversine==2.9.0
+jupyter_server==2.12.5
+geojson==3.2.0
+botocore==1.42.70
+fury==0.12.0
+ipympl==0.10.0
+ipython_pygments_lexers==1.1.1
+olefile==0.47
+jupyter_server_proxy==4.4.0
+datasets==4.8.3
+pytorch-ignite==0.5.3
+xvfbwrapper==0.2.22
+daal==2025.11.0
+open_spiel==1.6.12
+jupyter-lsp==1.5.1
+trx-python==0.4.0
+gpxpy==1.6.2
+papermill==2.7.0
+simpervisor==1.0.0
+kagglehub==1.0.0
+mlcrate==0.2.0
+kaggle==2.0.0
+dask-jobqueue==0.9.0
+model-signing==1.1.1
+jupyterlab==3.6.8
+args==0.1.0
+ImageHash==4.3.2
+typing-inspect==0.9.0
+PyUpSet==0.1.1.post7
+dacite==1.9.2
+pycryptodome==3.23.0
+google-cloud-videointelligence==2.18.0
+visions==0.8.1
+deap==1.4.3
+lml==0.2.0
+jiter==0.10.0
+ypy-websocket==0.8.4
+cytoolz==1.1.0
+path.py==12.5.0
+tensorflow-io==0.37.1
+wavio==0.0.9
+pdf2image==1.17.0
+line_profiler==5.0.2
+fsspec==2026.2.0
+aiobotocore==3.3.0
+optuna==4.8.0
+fastgit==0.0.4
+litellm==1.82.4
+pyLDAvis==3.4.1
+Janome==0.5.0
+langid==1.1.6
+sigstore-models==0.0.6
+pokerkit==0.6.3
+pyaml==26.2.1
+scikit-plot==0.3.7
+nbdev==3.0.12
+simpleitk==2.5.3
+ml_collections==1.1.0
+filetype==1.2.0
+Wand==0.7.0
+jupyter_server_ydoc==0.8.0
+pyjson5==2.0.0
+email-validator==2.3.0
+execnb==0.1.18
+colorama==0.4.6
+ruamel.yaml==0.19.1
+python-lsp-server==1.14.0
+black==26.3.1
+PyArabic==0.6.15
+gymnasium==1.2.0
+path==17.1.1
+gensim==4.4.0
+pypdf==6.9.1
+TPOT==1.1.0
+Pympler==1.1
+bayesian-optimization==3.2.1
+nbconvert==6.4.5
+kornia==0.8.2
+pathspec==1.0.4
+pybind11==3.0.2
+sigstore==4.2.0
+funcy==2.0
+func_timeout==4.3.5
+testpath==0.6.0
+aioitertools==0.13.0
+google-cloud-vision==3.12.1
+ray==2.54.0
+kornia_rs==0.1.10
+traitlets==5.14.3
+gymnax==0.0.8
+dnspython==2.8.0
+chex==0.1.90
+gym==0.26.2
+nbclient==0.5.13
+ydata-profiling==4.18.1
+POT==0.9.6.post1
+deepdiff==8.6.2
+squarify==0.4.4
+dataclasses-json==0.6.7
+pettingzoo==1.24.0
+pytorch-lightning==2.6.1
+segment_anything==1.0
+emoji==2.15.0
+python-bidi==0.6.7
+rgf-python==3.12.0
+ninja==1.13.0
+widgetsnbextension==4.0.15
+minify_html==0.18.1
+urwid==3.0.5
+jedi==0.19.2
+jupyterlab-lsp==3.10.2
+python-lsp-jsonrpc==1.1.2
+QtPy==2.4.3
+pydicom==3.0.1
+multimethod==1.12
+torchmetrics==1.9.0
+asttokens==3.0.1
+docker==7.1.0
+dask-expr==2.0.0
+s3transfer==0.16.0
+build==1.4.0
+Shimmy==2.0.0
+igraph==1.0.0
+puremagic==2.1.0
+jupyterlab_server==2.28.0
+isoweek==1.3.3
+texttable==1.7.0
+kt-legacy==1.0.5
+orderly-set==5.5.0
+pyexcel-io==0.6.7
+catboost==1.2.10
+kagglesdk==0.1.16
+mamba==0.11.3
+dipy==1.12.0
+colorlog==6.10.1
+asn1crypto==1.5.1
+pyexcel-ods==0.6.0
+lime==0.2.0.1
+pox==0.3.7
+rfc8785==0.1.4
+sigstore-rekor-types==0.0.18
+cesium==0.12.4
+boto3==1.42.70
+tuf==6.0.0
+hep_ml==0.8.0
+pyproject_hooks==1.2.0
+phik==0.12.5
+pudb==2025.1.5
+mne==1.11.0
+keras-cv==0.9.0
+dill==0.4.1
+gatspy==0.3
+scikit-learn-intelex==2025.11.0
+onnx==1.20.1
+scikit-optimize==0.10.2
+mypy_extensions==1.1.0
+mistune==0.8.4
+json5==0.13.0
+google-colab==1.0.0
+psutil==5.9.5
+jsonschema==4.26.0
+astunparse==1.6.3
+pycocotools==2.0.11
+lxml==6.0.2
+ipython==7.34.0
+oauthlib==3.3.1
+grpc-google-iam-v1==0.14.3
+array_record==0.8.3
+PuLP==3.3.0
+nvidia-cuda-runtime-cu12==12.8.90
+dask-cuda==26.2.0
+immutabledict==4.3.1
+peewee==4.0.0
+fiona==1.10.1
+aiosignal==1.4.0
+libclang==18.1.1
+annotated-types==0.7.0
+spreg==1.8.5
+grain==0.2.15
+geemap==0.35.3
+patsy==1.0.2
+imagesize==1.4.1
+py-cpuinfo==9.0.0
+pyzmq==26.2.1
+nvidia-cufile-cu12==1.13.1.3
+multidict==6.7.1
+srsly==2.5.2
+intel-openmp==2025.3.2
+uuid_utils==0.14.1
+google-cloud-language==2.19.0
+soxr==1.0.0
+jupyterlab_pygments==0.3.0
+backcall==0.2.0
+tensorflow-hub==0.16.1
+google==3.0.0
+requests-oauthlib==2.0.0
+dopamine_rl==4.1.2
+overrides==7.7.0
+db-dtypes==1.5.0
+jeepney==0.9.0
+langgraph-sdk==0.3.9
+ipython-genutils==0.2.0
+nvidia-cuda-cupti-cu12==12.8.90
+libcugraph-cu12==26.2.0
+catalogue==2.0.10
+beautifulsoup4==4.13.5
+nvidia-ml-py==13.590.48
+sphinxcontrib-devhelp==2.0.0
+partd==1.4.2
+sklearn-pandas==2.2.0
+sphinxcontrib-qthelp==2.0.0
+google-cloud-spanner==3.63.0
+h5py==3.15.1
+python-box==7.4.1
+distributed-ucxx-cu12==0.48.0
+xlrd==2.0.2
+branca==0.8.2
+chardet==5.2.0
+pycairo==1.29.0
+Authlib==1.6.8
+cuda-core==0.3.2
+sentencepiece==0.2.1
+nvidia-cusparselt-cu12==0.7.1
+matplotlib-venn==1.1.2
+scooby==0.11.0
+fqdn==1.5.1
+gin-config==0.5.0
+ipython-sql==0.5.0
+toml==0.10.2
+PyOpenGL==3.1.10
+weasel==0.4.3
+jsonpointer==3.0.0
+google-auth-httplib2==0.3.0
+spint==1.0.7
+nvtx==0.2.14
+websocket-client==1.9.0
+torchao==0.10.0
+splot==1.1.7
+langgraph-checkpoint==4.0.0
+alabaster==1.0.0
+jaxlib==0.7.2
+google-resumable-media==2.8.0
+namex==0.1.0
+quantecon==0.11.0
+nvidia-cuda-cccl-cu12==12.9.27
+google-cloud-aiplatform==1.138.0
+treelite==4.6.1
+google-cloud-resource-manager==1.16.0
+jupyter_core==5.9.1
+spacy-legacy==3.0.12
+librosa==0.11.0
+ibis-framework==9.5.0
+requests-toolbelt==1.0.0
+smart_open==7.5.1
+tensorflow-metadata==1.17.3
+pysal==25.7
+highspy==1.13.1
+click==8.3.1
+markdown-it-py==4.0.0
+nvidia-cusolver-cu12==11.7.3.90
+cupy-cuda12x==14.0.1
+imutils==0.5.4
+grpclib==0.4.9
+opt_einsum==3.4.0
+folium==0.20.0
+moviepy==1.0.3
+opencv-python==4.13.0.92
+en_core_web_sm==3.8.0
+tensorflow-text==2.19.0
+langchain-core==1.2.15
+yarl==1.22.0
+spacy==3.8.11
+importlib_resources==6.5.2
+peft==0.18.1
+lazy_loader==0.4
+polars-runtime-32==1.35.2
+pylibcudf-cu12==26.2.1
+bigquery-magics==0.10.3
+spanner-graph-notebook==1.1.8
+sqlglot==25.20.2
+linkify-it-py==2.0.3
+types-pytz==2025.2.0.20251108
+tifffile==2026.2.20
+tsfresh==0.21.1
+nbclassic==1.3.3
+scikit-image==0.25.2
+tensorflow_decision_forests==1.12.0
+simsimd==6.5.13
+isoduration==20.11.0
+momepy==0.11.0
+pytest==8.4.2
+nvidia-cuda-nvcc-cu12==12.5.82
+cuda-bindings==12.9.4
+torchsummary==1.5.1
+earthengine-api==1.5.24
+webencodings==0.5.1
+optree==0.19.0
+jax-cuda12-pjrt==0.7.2
+langchain==1.2.10
+safehttpx==0.1.7
+holidays==0.91
+google-cloud-firestore==2.23.0
+fastjsonschema==2.21.2
+pymc==5.28.0
+pydantic==2.12.3
+jaraco.context==6.1.0
+pyogrio==0.12.1
+numba-cuda==0.22.2
+fonttools==4.61.1
+httpimport==1.4.1
+rsa==4.9.1
+tomlkit==0.13.3
+entrypoints==0.4
+anyio==4.12.1
+charset-normalizer==3.4.4
+pooch==1.9.0
+libcuml-cu12==26.2.0
+astropy-iers-data==0.2026.2.23.0.48.33
+ipyleaflet==0.20.0
+cryptography==43.0.3
+missingno==0.5.2
+langgraph==1.0.9
+pandas-datareader==0.10.0
+pyviz_comms==3.0.6
+cycler==0.12.1
+tensorboard==2.19.0
+gast==0.7.0
+jax-cuda12-plugin==0.7.2
+platformdirs==4.9.2
+google-genai==1.64.0
+inflect==7.5.0
+httplib2==0.31.2
+h11==0.16.0
+alembic==1.18.4
+multitasking==0.0.12
+rmm-cu12==26.2.0
+cvxpy==1.6.7
+affine==2.4.0
+cuml-cu12==26.2.0
+pyparsing==3.3.2
+cffi==2.0.0
+h5netcdf==1.8.1
+Markdown==3.10.2
+google-cloud-translate==3.24.0
+rpy2==3.5.17
+regex==2025.11.3
+tf_keras==2.19.0
+google-auth==2.47.0
+nvidia-libnvcomp-cu12==5.1.0.21
+Send2Trash==2.1.0
+cymem==2.0.13
+pylibraft-cu12==26.2.0
+shap==0.50.0
+shapely==2.1.2
+psygnal==0.15.1
+uri-template==1.3.0
+parso==0.8.6
+webcolors==25.10.0
+nltk==3.9.1
+atpublic==5.1
+ImageIO==2.37.2
+sphinxcontrib-applehelp==2.0.0
+bigframes==2.35.0
+pydot==4.0.1
+onemkl-license==2025.3.1
+treescope==0.1.10
+tcmlib==1.4.1
+opentelemetry-sdk==1.38.0
+tiktoken==0.12.0
+nibabel==5.3.3
+multiprocess==0.70.16
+typing_extensions==4.15.0
+PyYAML==6.0.3
+defusedxml==0.7.1
+sphinxcontrib-serializinghtml==2.0.0
+bleach==6.3.0
+tenacity==9.1.4
+python-utils==3.9.1
+google-cloud-bigquery==3.40.1
+google-cloud-bigquery-connection==1.20.0
+opentelemetry-resourcedetector-gcp==1.11.0a0
+ormsgpack==1.12.2
+pydotplus==2.0.2
+pycryptodomex==3.23.0
+openai==2.23.0
+matplotlib==3.10.0
+ml_dtypes==0.5.4
+uvloop==0.22.1
+google-pasta==0.2.0
+giddy==2.3.8
+ipyparallel==8.8.0
+keras==3.10.0
+cuvs-cu12==26.2.0
+mcp==1.26.0
+spacy-loggers==1.0.5
+google-cloud-logging==3.13.0
+rfc3987-syntax==1.1.0
+google-ai-generativelanguage==0.6.15
+keras-hub==0.21.1
+pydata-google-auth==1.9.1
+absl-py==1.4.0
+ydf==0.15.0
+narwhals==2.17.0
+nvidia-cusparse-cu12==12.5.8.93
+openpyxl==3.1.5
+nvidia-cublas-cu12==12.8.4.1
+roman-numerals==4.1.0
+vega-datasets==0.9.0
+mpmath==1.3.0
+etils==1.13.0
+osqp==1.1.1
+traittypes==0.2.3
+opentelemetry-exporter-gcp-monitoring==1.11.0a0
+graphviz==0.21
+google-cloud-trace==1.18.0
+einops==0.8.2
+torchdata==0.11.0
+jax==0.7.2
+cachetools==6.2.6
+aiohappyeyeballs==2.6.1
+annotated-doc==0.0.4
+starlette==0.52.1
+fastapi==0.133.0
+typer==0.24.1
+duckdb==1.3.2
+blinker==1.9.0
+referencing==0.37.0
+googledrivedownloader==1.1.0
+GDAL==3.8.4
+cuda-python==12.9.4
+pycparser==3.0
+et_xmlfile==2.0.0
+jieba==0.42.1
+zict==3.0.0
+hyperopt==0.2.7
+python-louvain==0.16
+SQLAlchemy==2.0.47
+cuda-toolkit==12.8.1
+PyDrive2==1.21.3
+roman-numerals-py==4.1.0
+urllib3==2.5.0
+jaraco.functools==4.4.0
+optax==0.2.7
+pyOpenSSL==24.2.1
+jupyter-console==6.6.3
+libkvikio-cu12==26.2.0
+gspread==6.2.1
+docstring_parser==0.17.0
+albumentations==2.0.8
+jupytext==1.19.1
+seaborn==0.13.2
+librmm-cu12==26.2.0
+cons==0.4.7
+matplotlib-inline==0.2.1
+pynndescent==0.6.0
+stringzilla==4.6.0
+flatbuffers==25.12.19
+omegaconf==2.3.0
+umap-learn==0.5.11
+progressbar2==4.5.0
+pexpect==4.9.0
+torchcodec==0.10.0+cu128
+ptyprocess==0.7.0
+pygame==2.6.1
+kiwisolver==1.4.9
+Cython==3.0.12
+shellingham==1.5.4
+soupsieve==2.8.3
+snowballstemmer==3.0.1
+propcache==0.4.1
+ucxx-cu12==0.48.0
+nbformat==5.10.4
+python-snappy==0.7.3
+rasterstats==0.20.0
+bqplot==0.12.45
+nest-asyncio==1.6.0
+opencv-python-headless==4.13.0.92
+notebook==6.5.7
+flax==0.11.2
+google-cloud-functions==1.22.0
+multipledispatch==1.0.0
+googleapis-common-protos==1.72.0
+eerepr==0.1.2
+torchaudio==2.10.0+cu128
+locket==1.0.0
+prettytable==3.17.0
+pygit2==1.19.1
+fastai==2.8.7
+msgpack==1.1.2
+clarabel==0.11.1
+cligj==0.7.2
+google-cloud-secret-manager==2.26.0
+spglm==1.1.0
+ipytree==0.2.2
+termcolor==3.3.0
+tweepy==4.16.0
+google-cloud-core==2.5.0
+dataproc-spark-connect==1.0.2
+mkl==2025.3.1
+umf==1.0.3
+textblob==0.19.0
+firebase-admin==6.9.0
+simple-parsing==0.1.8
+debugpy==1.8.15
+google-cloud-discoveryengine==0.13.12
+fastcore==1.12.16
+decorator==4.4.2
+pickleshare==0.7.5
+rasterio==1.5.0
+networkx==3.6.1
+typer-slim==0.24.0
+wasabi==1.1.3
+mgwr==2.2.1
+hdbscan==0.8.41
+pydub==0.25.1
+tobler==0.13.0
+more-itertools==10.8.0
+keyrings.google-artifactregistry-auth==1.1.2
+cloudpickle==3.1.2
+nvidia-nvtx-cu12==12.8.90
+fastlite==0.2.4
+colorcet==3.1.0
+lark==1.3.1
+antlr4-python3-runtime==4.9.3
+keras-nlp==0.21.1
+music21==9.9.1
+Pygments==2.19.2
+triton==3.6.0
+toolz==0.12.1
+python-slugify==8.0.4
+sqlparse==0.5.5
+jupyter-leaflet==0.20.0
+gym-notices==0.1.0
+torchvision==0.25.0+cu128
+prophet==1.3.0
+google-cloud-datastore==2.23.0
+semantic-version==2.10.0
+fastprogress==1.1.5
+etuples==0.3.10
+pyspark==4.0.2
+orjson==3.11.7
+terminado==0.18.1
+accelerate==1.12.0
+panel==1.8.7
+apswutils==0.1.2
+pyproj==3.7.2
+sphinxcontrib-htmlhelp==2.1.0
+certifi==2026.1.4
+grpc-interceptor==0.15.4
+pyasn1==0.6.2
+geocoder==1.38.1
+idna==3.11
+mizani==0.13.5
+jupyter_server_terminals==0.5.4
+httpcore==1.0.9
+pyasn1_modules==0.4.2
+ffmpy==1.0.0
+pyperclip==1.11.0
+safetensors==0.7.0
+ndindex==1.10.1
+tblib==3.2.2
+docutils==0.21.2
+scs==3.2.11
+distro==1.9.0
+tf-slim==1.1.0
+babel==2.18.0
+google-cloud-pubsub==2.35.0
+google-api-python-client==2.190.0
+tzlocal==5.3.1
+groovy==0.1.2
+plum-dispatch==2.7.1
+dask==2026.1.1
+blosc2==4.0.0
+sqlalchemy-spanner==1.17.2
+orbax-checkpoint==0.11.33
+wandb==0.25.0
+geopandas==1.1.2
+proglog==0.1.12
+python-dateutil==2.9.0.post0
+tzdata==2025.3
+editdistance==0.8.1
+langsmith==0.7.6
+xarray-einstats==0.10.0
+pydantic_core==2.41.4
+tabulate==0.9.0
+mmh3==5.2.0
+sentry-sdk==2.53.0
+spopt==0.7.0
+dlib==19.24.6
+community==1.0.0b1
+tensorflow==2.19.0
+ale-py==0.11.2
+murmurhash==1.0.15
+notebook_shim==0.2.4
+mdurl==0.1.2
+diffusers==0.36.0
+requests==2.32.4
+Flask==3.1.3
+prometheus_client==0.24.1
+uvicorn==0.41.0
+logical-unification==0.4.7
+soundfile==0.13.1
+itsdangerous==2.2.0
+jsonpatch==1.33
+plotnine==0.14.5
+distributed==2026.1.1
+google-auth-oauthlib==1.2.4
+gdown==5.2.1
+brotli==1.2.0
+py4j==0.10.9.9
+pytensor==2.38.0
+text-unidecode==1.3
+yfinance==0.2.66
+arviz==0.22.0
+cudf-cu12==26.2.1
+wordcloud==1.9.6
+jaraco.classes==3.4.0
+albucore==0.0.24
+python-dotenv==1.2.1
+uritemplate==4.2.0
+nx-cugraph-cu12==26.2.0
+raft-dask-cu12==26.2.0
+hpack==4.1.0
+numexpr==2.14.1
+pydantic-settings==2.13.1
+rapids-logger==0.2.3
+cmake==3.31.10
+pillow==11.3.0
+jsonschema-specifications==2025.9.1
+tables==3.10.2
+google-cloud-storage==3.9.0
+mapclassify==2.10.0
+altair==5.5.0
+filelock==3.24.3
+google-cloud-appengine-logging==1.8.0
+cufflinks==0.17.3
+cvxopt==1.3.2
+six==1.17.0
+watchdog==6.0.0
+sse-starlette==3.2.0
+PySocks==1.7.1
+jupyterlab_widgets==3.0.16
+spaghetti==1.7.6
+intel-cmplr-lib-ur==2025.3.2
+uc-micro-py==1.0.3
+Sphinx==8.2.3
+PyJWT==2.11.0
+google-cloud-bigtable==2.35.0
+numba==0.60.0
+httptools==0.7.1
+rich==13.9.4
+pointpats==2.5.5
+watchfiles==1.1.1
+promise==2.3
+polars==1.35.2
+greenlet==3.3.2
+rfc3986-validator==0.1.1
+threadpoolctl==3.6.0
+opentelemetry-exporter-otlp-proto-http==1.38.0
+libcuvs-cu12==26.2.0
+sniffio==1.3.1
+pylibcugraph-cu12==26.2.0
+holoviews==1.22.1
+pandas-gbq==0.30.0
+frozenlist==1.8.0
+google-crc32c==1.8.0
+torch==2.10.0+cu128
+ipyevents==2.0.4
+libucxx-cu12==0.48.0
+cramjam==2.11.0
+opentelemetry-exporter-otlp-proto-common==1.38.0
+wurlitzer==3.1.1
+confection==0.1.5
+stanio==0.5.1
+easydict==1.13
+argon2-cffi==25.1.0
+llvmlite==0.43.0
+humanize==4.15.0
+rapids-dask-dependency==26.2.0
+argon2-cffi-bindings==25.1.0
+future==1.0.0
+rpds-py==0.30.0
+psycopg2==2.9.11
+iniconfig==2.3.0
+jupyter-events==0.12.0
+nvidia-nccl-cu12==2.27.5
+GitPython==3.1.46
+joblib==1.5.3
+beartype==0.22.9
+hf-xet==1.3.0
+Bottleneck==1.4.2
+apsw==3.51.2.0
+bokeh==3.8.2
+google-cloud-dataproc==5.25.0
+nvidia-cuda-nvrtc-cu12==12.8.93
+colour==0.1.5
+zipp==3.23.0
+blis==1.3.3
+click-plugins==1.1.1.2
+httpx-sse==0.4.3
+nvidia-nvshmem-cu12==3.4.5
+sphinxcontrib-jsmath==1.0.1
+prompt_toolkit==3.0.52
+esda==2.8.1
+param==2.3.2
+google-cloud-speech==2.36.1
+portpicker==1.5.2
+PyWavelets==1.9.0
+google-cloud-monitoring==2.29.1
+Farama-Notifications==0.0.4
+pytz==2025.2
+MarkupSafe==3.0.3
+pyomo==6.10.0
+packaging==26.0
+betterproto==2.0.0b6
+libraft-cu12==26.2.0
+typeguard==4.5.1
+imbalanced-learn==0.14.1
+google-adk==1.25.1
+CacheControl==0.14.4
+ipykernel==6.17.1
+jsonpickle==4.1.1
+xyzservices==2025.11.0
+websockets==15.0.1
+PyGObject==3.48.2
+pandas-stubs==2.2.2.240909
+proto-plus==1.27.1
+segregation==2.5.3
+ratelim==0.1.6
+miniKanren==1.0.5
+geographiclib==2.1
+Jinja2==3.1.6
+frozendict==2.4.7
+libcudf-cu12==26.2.1
+nvidia-cufft-cu12==11.3.3.83
+typing-inspection==0.4.2
+gradio_client==1.14.0
+simplejson==3.20.2
+ruff==0.15.2
+imageio-ffmpeg==0.6.0
+python-json-logger==4.0.0
+cucim-cu12==26.2.0
+jupyter_kernel_gateway==2.5.2
+contourpy==1.3.3
+google-api-core==2.30.0
+opencv-contrib-python==4.13.0.92
+nvidia-cudnn-cu12==9.10.2.21
+opentelemetry-proto==1.38.0
+dask-cudf-cu12==26.2.1
+nvidia-nvimgcodec-cu12==0.7.0.11
+statsmodels==0.14.6
+opentelemetry-exporter-gcp-trace==1.11.0
+deprecation==2.1.0
+tinycss2==1.4.0
+mdit-py-plugins==0.5.0
+tensorflow-datasets==4.9.9
+opentelemetry-api==1.38.0
+langgraph-prebuilt==1.0.8
+keyring==25.7.0
+inequality==1.1.2
+cyipopt==1.5.0
+sympy==1.14.0
+oauth2client==4.1.3
+python-fasthtml==0.12.47
+gspread-dataframe==4.0.0
+wcwidth==0.6.0
+geopy==2.4.1
+natsort==8.4.0
+timm==1.0.25
+rfc3339-validator==0.1.4
+stumpy==1.13.0
+parsy==2.2
+libucx-cu12==1.19.0
+pyerfa==2.0.1.5
+astropy==7.2.0
+curl_cffi==0.14.0
+xarray==2025.12.0
+preshed==3.0.12
+Werkzeug==3.1.6
+SecretStorage==3.5.0
+grpcio==1.78.1
+slicer==0.0.8
+cudf-polars-cu12==26.2.1
+aiosqlite==0.22.1
+grpcio-status==1.71.2
+libpysal==4.14.1
+gitdb==4.0.12
+hyperframe==6.1.0
+opentelemetry-semantic-conventions==0.59b0
+wheel==0.46.3
+h2==4.3.0
+google-cloud-audit-log==0.4.0
+tqdm==4.67.3
+httpx==0.28.1
+cloudpathlib==0.23.0
+thinc==8.3.10
+audioread==3.1.0
+fastdownload==0.0.7
+gcsfs==2025.3.0
+nvidia-nvjitlink-cu12==12.8.93
+access==1.1.10.post3
+tornado==6.5.1
+pandocfilters==1.5.1
+fasttransform==0.0.2
+nvidia-curand-cu12==10.3.9.90
+python-multipart==0.0.22
+yellowbrick==1.5
+jupyter_client==7.4.9
+google-generativeai==0.8.6
+blobfile==3.2.0
+importlib_metadata==8.7.1
+tensorboard-data-server==0.7.2
+attrs==25.4.0
+tbb==2022.3.1
+pluggy==1.6.0
+cuda-pathfinder==1.3.5
+rtree==1.4.1
+arrow==1.4.0
+wrapt==2.1.1
+anywidget==0.9.21
+mlxtend==0.23.4
+smmap==5.0.2
+aiohttp==3.13.3
+opentelemetry-exporter-gcp-logging==1.11.0a0
+sortedcontainers==2.4.0
+pyshp==3.0.3
+sklearn-compat==0.1.5
+xxhash==3.6.0
+zstandard==0.25.0
+Mako==1.3.10
+google-cloud-iam==2.21.0
+autograd==1.8.0
+glob2==0.7
+tensorstore==0.1.81
+tensorflow-probability==0.25.0
+colorlover==0.3.0
+ipyfilechooser==0.6.0
+gradio==5.50.0
+cmdstanpy==1.3.0
+dm-tree==0.1.9
+html5lib==1.1
+python-apt==0.0.0
+PyGObject==3.42.1
+blinker==1.4
+jeepney==0.7.1
+six==1.16.0
+oauthlib==3.2.0
+wadllib==1.3.6
+launchpadlib==1.10.16
+dbus-python==1.2.18
+PyJWT==2.3.0
+importlib-metadata==4.6.4
+httplib2==0.20.2
+zipp==1.0.0
+pyparsing==2.4.7
+lazr.restfulclient==0.14.4
+SecretStorage==3.3.1
+distro==1.7.0
+lazr.uri==1.0.6
+more-itertools==8.10.0
+python-apt==2.4.0+ubuntu4.1
+cryptography==3.4.8
+keyring==23.5.0
+Markdown==3.3.6
+Mako==1.1.3
+MarkupSafe==2.0.1
+packaging==24.1
+inflect==7.3.1
+autocommand==2.2.2
+typeguard==4.3.0
+jaraco.text==3.12.1
+importlib_resources==6.4.0
+wheel==0.43.0
+zipp==3.19.2
+platformdirs==4.2.2
+importlib_metadata==8.0.0
+tomli==2.0.1
+jaraco.collections==5.1.0
+more-itertools==10.3.0
+typing_extensions==4.12.2
+backports.tarfile==1.2.0
+jaraco.functools==4.0.1
+jaraco.context==5.3.0

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/files/wandb-metadata.json

@@ -0,0 +1,42 @@
+{
+  "os":  "Linux-6.6.113+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.12.12",
+  "startedAt":  "2026-03-31T09:47:28.010618Z",
+  "program":  "kaggle.ipynb",
+  "codePath":  "kaggle.ipynb",
+  "email":  "suvraadeep@gmail.com",
+  "root":  "/kaggle/working",
+  "host":  "7a129c53b2b5",
+  "executable":  "/usr/bin/python3",
+  "cpu_count":  2,
+  "cpu_count_logical":  4,
+  "gpu":  "Tesla T4",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "8656922775552",
+      "used":  "7347653709824"
+    }
+  },
+  "memory":  {
+    "total":  "33662472192"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "Tesla T4",
+      "memoryTotal":  "16106127360",
+      "cudaCores":  2560,
+      "architecture":  "Turing",
+      "uuid":  "GPU-c7ebff42-1f98-33a6-9169-a1e3925c4f52"
+    },
+    {
+      "name":  "Tesla T4",
+      "memoryTotal":  "16106127360",
+      "cudaCores":  2560,
+      "architecture":  "Turing",
+      "uuid":  "GPU-4f057d35-8a41-ce55-5d7d-60c77ebb7156"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "s21p4sab0yaskqym8r37eeuqwjpzso9d"
+}

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":653},"final_oof_auc":0.7873122800363673,"lgbm_weight":0.49,"xgb_weight":0.51,"_timestamp":1.7749511069771621e+09,"_step":0,"_runtime":653}

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/logs/debug-internal.log

@@ -0,0 +1,12 @@
+{"time":"2026-03-31T09:47:28.022227997Z","level":"INFO","msg":"stream: starting","core version":"0.25.0"}
+{"time":"2026-03-31T09:47:28.143809348Z","level":"INFO","msg":"stream: created new stream","id":"v7xaa9j7"}
+{"time":"2026-03-31T09:47:28.143904318Z","level":"INFO","msg":"handler: started","stream_id":"v7xaa9j7"}
+{"time":"2026-03-31T09:47:28.144018966Z","level":"INFO","msg":"stream: started","id":"v7xaa9j7"}
+{"time":"2026-03-31T09:47:28.144050221Z","level":"INFO","msg":"writer: started","stream_id":"v7xaa9j7"}
+{"time":"2026-03-31T09:47:28.144080736Z","level":"INFO","msg":"sender: started","stream_id":"v7xaa9j7"}
+{"time":"2026-03-31T09:58:27.365155501Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-03-31T09:58:27.509723149Z","level":"INFO","msg":"handler: operation stats","stats":{}}
+{"time":"2026-03-31T09:58:27.515863816Z","level":"INFO","msg":"stream: closing","id":"v7xaa9j7"}
+{"time":"2026-03-31T09:58:27.515880856Z","level":"INFO","msg":"handler: closed","stream_id":"v7xaa9j7"}
+{"time":"2026-03-31T09:58:27.516043135Z","level":"INFO","msg":"sender: closed","stream_id":"v7xaa9j7"}
+{"time":"2026-03-31T09:58:27.51608442Z","level":"INFO","msg":"stream: closed","id":"v7xaa9j7"}

kaggle_output/wandb/run-20260331_094728-v7xaa9j7/logs/debug.log

@@ -0,0 +1,26 @@
+2026-03-31 09:47:28,019 INFO    MainThread:177 [wandb_init.py:setup_run_log_directory():717] Logging user logs to /kaggle/working/wandb/run-20260331_094728-v7xaa9j7/logs/debug.log
+2026-03-31 09:47:28,019 INFO    MainThread:177 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to /kaggle/working/wandb/run-20260331_094728-v7xaa9j7/logs/debug-internal.log
+2026-03-31 09:47:28,019 INFO    MainThread:177 [wandb_init.py:monkeypatch_ipython():636] configuring jupyter hooks <wandb.sdk.wandb_init._WandbInit object at 0x7e0be8427da0>
+2026-03-31 09:47:28,020 INFO    MainThread:177 [wandb_init.py:init():844] calling init triggers
+2026-03-31 09:47:28,020 INFO    MainThread:177 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
+config: {'lgbm': {'boosting_type': 'gbdt', 'num_leaves': 112, 'learning_rate': 0.02273805573563183, 'feature_fraction': 0.7123738038749523, 'bagging_fraction': 0.8170784332632994, 'bagging_freq': 1, 'min_child_samples': 82, 'reg_alpha': 0.0002359137306347715, 'reg_lambda': 8.598737339212267, 'max_depth': 9, 'min_gain_to_split': 0.0993578407670862, 'objective': 'binary', 'metric': 'auc', 'verbosity': -1, 'n_jobs': -1, 'seed': 42, 'device': 'gpu', 'max_bin': 255, 'gpu_use_dp': False, 'force_col_wise': True}, 'xgb': {'learning_rate': 0.028084447839949865, 'max_depth': 5, 'min_child_weight': 4, 'subsample': 0.6252365594243677, 'colsample_bytree': 0.8226259232371986, 'gamma': 0.26516420723672285, 'reg_alpha': 8.576808828106026, 'reg_lambda': 6.598703417106125, 'scale_pos_weight': 1.3690073364184308, 'objective': 'binary:logistic', 'eval_metric': 'auc', 'tree_method': 'hist', 'seed': 42, 'n_jobs': -1, 'device': 'cuda', 'max_bin': 256}, 'n_folds': 5, '_wandb': {}}
+2026-03-31 09:47:28,020 INFO    MainThread:177 [wandb_init.py:init():892] starting backend
+2026-03-31 09:47:28,020 INFO    MainThread:177 [wandb_init.py:init():895] sending inform_init request
+2026-03-31 09:47:28,021 INFO    MainThread:177 [wandb_init.py:init():903] backend started and connected
+2026-03-31 09:47:28,029 INFO    MainThread:177 [wandb_run.py:_label_probe_notebook():1333] probe notebook
+2026-03-31 09:47:33,411 INFO    MainThread:177 [wandb_init.py:init():973] updated telemetry
+2026-03-31 09:47:33,412 INFO    MainThread:177 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
+2026-03-31 09:47:33,771 INFO    MainThread:177 [wandb_init.py:init():1042] starting run threads in backend
+2026-03-31 09:47:34,481 INFO    MainThread:177 [wandb_run.py:_console_start():2524] atexit reg
+2026-03-31 09:47:34,481 INFO    MainThread:177 [wandb_run.py:_redirect():2373] redirect: wrap_raw
+2026-03-31 09:47:34,481 INFO    MainThread:177 [wandb_run.py:_redirect():2442] Wrapping output streams.
+2026-03-31 09:47:34,482 INFO    MainThread:177 [wandb_run.py:_redirect():2465] Redirects installed.
+2026-03-31 09:47:34,484 INFO    MainThread:177 [wandb_init.py:init():1082] run started, returning control to user process
+2026-03-31 09:58:26,977 INFO    MainThread:177 [wandb_run.py:_finish():2291] finishing run suvradeep-iit-guwahati-/credit-invisibility/v7xaa9j7
+2026-03-31 09:58:26,978 INFO    MainThread:177 [jupyter.py:save_history():435] not saving jupyter history
+2026-03-31 09:58:26,978 INFO    MainThread:177 [jupyter.py:save_ipynb():362] not saving jupyter notebook
+2026-03-31 09:58:26,978 INFO    MainThread:177 [wandb_init.py:_jupyter_teardown():621] cleaning up jupyter logic
+2026-03-31 09:58:26,978 INFO    MainThread:177 [wandb_run.py:_atexit_cleanup():2490] got exitcode: 0
+2026-03-31 09:58:26,978 INFO    MainThread:177 [wandb_run.py:_restore():2472] restore
+2026-03-31 09:58:26,979 INFO    MainThread:177 [wandb_run.py:_restore():2478] restore done
+2026-03-31 09:58:27,513 INFO    MainThread:177 [wandb_run.py:_footer_sync_info():3868] logging synced files

kaggle_output/wandb/run-20260331_095935-cu87492i/files/config.yaml

@@ -0,0 +1,85 @@
+_wandb:
+    value:
+        cli_version: 0.25.0
+        e:
+            8a2fdkg7z7uti6wkzjkth5zruzlcpawp:
+                codePath: kaggle.ipynb
+                cpu_count: 2
+                cpu_count_logical: 4
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "8656922775552"
+                        used: "7347718606848"
+                email: suvraadeep@gmail.com
+                executable: /usr/bin/python3
+                gpu: Tesla T4
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Turing
+                      cudaCores: 2560
+                      memoryTotal: "16106127360"
+                      name: Tesla T4
+                      uuid: GPU-c7ebff42-1f98-33a6-9169-a1e3925c4f52
+                    - architecture: Turing
+                      cudaCores: 2560
+                      memoryTotal: "16106127360"
+                      name: Tesla T4
+                      uuid: GPU-4f057d35-8a41-ce55-5d7d-60c77ebb7156
+                host: 7a129c53b2b5
+                memory:
+                    total: "33662472192"
+                os: Linux-6.6.113+-x86_64-with-glibc2.35
+                program: kaggle.ipynb
+                python: CPython 3.12.12
+                root: /kaggle/working
+                startedAt: "2026-03-31T09:59:35.665971Z"
+                writerId: 8a2fdkg7z7uti6wkzjkth5zruzlcpawp
+        m: []
+        python_version: 3.12.12
+        t:
+            "1":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 6
+                - 8
+                - 11
+                - 12
+                - 35
+                - 49
+                - 53
+                - 54
+                - 71
+                - 75
+                - 105
+            "2":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 6
+                - 8
+                - 11
+                - 12
+                - 35
+                - 49
+                - 53
+                - 54
+                - 71
+                - 75
+                - 105
+            "3":
+                - 2
+                - 13
+                - 15
+            "4": 3.12.12
+            "5": 0.25.0
+            "6": 4.41.2
+            "8":
+                - 1
+                - 2
+                - 12
+            "12": 0.25.0
+            "13": linux-x86_64

kaggle_output/wandb/run-20260331_095935-cu87492i/files/output.log

@@ -0,0 +1,3 @@
+[LightGBM] [Fatal] The number of features in data (30) is not the same as it was in training data (234).
+You can set ``predict_disable_shape_check=true`` to discard this error, but please be aware what you are doing.
+📉 Simulating Concept Drift...

kaggle_output/wandb/run-20260331_095935-cu87492i/files/requirements.txt

@@ -0,0 +1,974 @@
+setuptools==75.2.0
+types-setuptools==80.10.0.20260124
+requirements-parser==0.9.0
+pip==24.1.2
+logistro==2.0.1
+tokenizers==0.19.1
+huggingface_hub==0.36.2
+scikit-learn==1.4.2
+xgboost==2.0.3
+sentence-transformers==2.7.0
+choreographer==1.2.1
+lightgbm==4.3.0
+pytest-timeout==2.4.0
+numpy==1.26.4
+category-encoders==2.6.3
+scipy==1.11.4
+river==0.23.0
+optuna-integration==4.8.0
+kaleido==1.2.0
+transformers==4.41.2
+plotly==6.6.0
+pytools==2025.2.5
+pycuda==2026.1
+siphash24==1.8
+protobuf==5.29.5
+torchtune==0.6.1
+learntools==0.3.5
+rouge_score==0.1.2
+pyclipper==1.4.0
+urwid_readline==0.15.1
+h2o==3.46.0.10
+rfc3161-client==1.0.5
+blake3==1.0.8
+mpld3==0.5.12
+qgrid==1.3.1
+ConfigSpace==1.2.2
+woodwork==0.31.0
+ujson==5.12.0
+y-py==0.6.2
+ipywidgets==8.1.5
+scikit-multilearn==0.2.0
+lightning-utilities==0.15.3
+pytesseract==0.3.13
+Cartopy==0.25.0
+odfpy==1.4.1
+Boruta==0.4.3
+docstring-to-markdown==0.17
+torchinfo==1.8.0
+clint==0.5.1
+comm==0.2.3
+Deprecated==1.3.1
+pymongo==4.16.0
+tensorflow-io-gcs-filesystem==0.37.1
+jmespath==1.1.0
+pygltflib==1.16.5
+keras-core==0.1.7
+pandas==2.3.3
+securesystemslib==1.3.1
+ghapi==1.0.11
+qtconsole==5.7.1
+pyemd==2.0.0
+pandas-profiling==3.6.6
+nilearn==0.13.1
+in-toto-attestation==0.9.3
+a2a-sdk==0.3.25
+keras-tuner==1.4.8
+fastuuid==0.14.0
+scikit-surprise==1.1.4
+vtk==9.3.1
+jupyter-ydoc==0.2.5
+aiofiles==22.1.0
+pytokens==0.4.1
+featuretools==1.31.0
+plotly-express==0.4.1
+marshmallow==3.26.2
+easyocr==1.7.2
+ppft==1.7.8
+openslide-bin==4.0.0.13
+fuzzywuzzy==0.18.0
+id==1.6.1
+openslide-python==1.4.3
+kaggle-environments==1.27.3
+pyarrow==23.0.1
+pandasql==0.7.3
+update-checker==0.18.0
+pathos==0.3.2
+jupyter_server_fileid==0.9.3
+fasttext==0.9.3
+coverage==7.13.5
+s3fs==2026.2.0
+stopit==1.1.2
+haversine==2.9.0
+jupyter_server==2.12.5
+geojson==3.2.0
+botocore==1.42.70
+fury==0.12.0
+ipympl==0.10.0
+ipython_pygments_lexers==1.1.1
+olefile==0.47
+jupyter_server_proxy==4.4.0
+datasets==4.8.3
+pytorch-ignite==0.5.3
+xvfbwrapper==0.2.22
+daal==2025.11.0
+open_spiel==1.6.12
+jupyter-lsp==1.5.1
+trx-python==0.4.0
+gpxpy==1.6.2
+papermill==2.7.0
+simpervisor==1.0.0
+kagglehub==1.0.0
+mlcrate==0.2.0
+kaggle==2.0.0
+dask-jobqueue==0.9.0
+model-signing==1.1.1
+jupyterlab==3.6.8
+args==0.1.0
+ImageHash==4.3.2
+typing-inspect==0.9.0
+PyUpSet==0.1.1.post7
+dacite==1.9.2
+pycryptodome==3.23.0
+google-cloud-videointelligence==2.18.0
+visions==0.8.1
+deap==1.4.3
+lml==0.2.0
+jiter==0.10.0
+ypy-websocket==0.8.4
+cytoolz==1.1.0
+path.py==12.5.0
+tensorflow-io==0.37.1
+wavio==0.0.9
+pdf2image==1.17.0
+line_profiler==5.0.2
+fsspec==2026.2.0
+aiobotocore==3.3.0
+optuna==4.8.0
+fastgit==0.0.4
+litellm==1.82.4
+pyLDAvis==3.4.1
+Janome==0.5.0
+langid==1.1.6
+sigstore-models==0.0.6
+pokerkit==0.6.3
+pyaml==26.2.1
+scikit-plot==0.3.7
+nbdev==3.0.12
+simpleitk==2.5.3
+ml_collections==1.1.0
+filetype==1.2.0
+Wand==0.7.0
+jupyter_server_ydoc==0.8.0
+pyjson5==2.0.0
+email-validator==2.3.0
+execnb==0.1.18
+colorama==0.4.6
+ruamel.yaml==0.19.1
+python-lsp-server==1.14.0
+black==26.3.1
+PyArabic==0.6.15
+gymnasium==1.2.0
+path==17.1.1
+gensim==4.4.0
+pypdf==6.9.1
+TPOT==1.1.0
+Pympler==1.1
+bayesian-optimization==3.2.1
+nbconvert==6.4.5
+kornia==0.8.2
+pathspec==1.0.4
+pybind11==3.0.2
+sigstore==4.2.0
+funcy==2.0
+func_timeout==4.3.5
+testpath==0.6.0
+aioitertools==0.13.0
+google-cloud-vision==3.12.1
+ray==2.54.0
+kornia_rs==0.1.10
+traitlets==5.14.3
+gymnax==0.0.8
+dnspython==2.8.0
+chex==0.1.90
+gym==0.26.2
+nbclient==0.5.13
+ydata-profiling==4.18.1
+POT==0.9.6.post1
+deepdiff==8.6.2
+squarify==0.4.4
+dataclasses-json==0.6.7
+pettingzoo==1.24.0
+pytorch-lightning==2.6.1
+segment_anything==1.0
+emoji==2.15.0
+python-bidi==0.6.7
+rgf-python==3.12.0
+ninja==1.13.0
+widgetsnbextension==4.0.15
+minify_html==0.18.1
+urwid==3.0.5
+jedi==0.19.2
+jupyterlab-lsp==3.10.2
+python-lsp-jsonrpc==1.1.2
+QtPy==2.4.3
+pydicom==3.0.1
+multimethod==1.12
+torchmetrics==1.9.0
+asttokens==3.0.1
+docker==7.1.0
+dask-expr==2.0.0
+s3transfer==0.16.0
+build==1.4.0
+Shimmy==2.0.0
+igraph==1.0.0
+puremagic==2.1.0
+jupyterlab_server==2.28.0
+isoweek==1.3.3
+texttable==1.7.0
+kt-legacy==1.0.5
+orderly-set==5.5.0
+pyexcel-io==0.6.7
+catboost==1.2.10
+kagglesdk==0.1.16
+mamba==0.11.3
+dipy==1.12.0
+colorlog==6.10.1
+asn1crypto==1.5.1
+pyexcel-ods==0.6.0
+lime==0.2.0.1
+pox==0.3.7
+rfc8785==0.1.4
+sigstore-rekor-types==0.0.18
+cesium==0.12.4
+boto3==1.42.70
+tuf==6.0.0
+hep_ml==0.8.0
+pyproject_hooks==1.2.0
+phik==0.12.5
+pudb==2025.1.5
+mne==1.11.0
+keras-cv==0.9.0
+dill==0.4.1
+gatspy==0.3
+scikit-learn-intelex==2025.11.0
+onnx==1.20.1
+scikit-optimize==0.10.2
+mypy_extensions==1.1.0
+mistune==0.8.4
+json5==0.13.0
+google-colab==1.0.0
+psutil==5.9.5
+jsonschema==4.26.0
+astunparse==1.6.3
+pycocotools==2.0.11
+lxml==6.0.2
+ipython==7.34.0
+oauthlib==3.3.1
+grpc-google-iam-v1==0.14.3
+array_record==0.8.3
+PuLP==3.3.0
+nvidia-cuda-runtime-cu12==12.8.90
+dask-cuda==26.2.0
+immutabledict==4.3.1
+peewee==4.0.0
+fiona==1.10.1
+aiosignal==1.4.0
+libclang==18.1.1
+annotated-types==0.7.0
+spreg==1.8.5
+grain==0.2.15
+geemap==0.35.3
+patsy==1.0.2
+imagesize==1.4.1
+py-cpuinfo==9.0.0
+pyzmq==26.2.1
+nvidia-cufile-cu12==1.13.1.3
+multidict==6.7.1
+srsly==2.5.2
+intel-openmp==2025.3.2
+uuid_utils==0.14.1
+google-cloud-language==2.19.0
+soxr==1.0.0
+jupyterlab_pygments==0.3.0
+backcall==0.2.0
+tensorflow-hub==0.16.1
+google==3.0.0
+requests-oauthlib==2.0.0
+dopamine_rl==4.1.2
+overrides==7.7.0
+db-dtypes==1.5.0
+jeepney==0.9.0
+langgraph-sdk==0.3.9
+ipython-genutils==0.2.0
+nvidia-cuda-cupti-cu12==12.8.90
+libcugraph-cu12==26.2.0
+catalogue==2.0.10
+beautifulsoup4==4.13.5
+nvidia-ml-py==13.590.48
+sphinxcontrib-devhelp==2.0.0
+partd==1.4.2
+sklearn-pandas==2.2.0
+sphinxcontrib-qthelp==2.0.0
+google-cloud-spanner==3.63.0
+h5py==3.15.1
+python-box==7.4.1
+distributed-ucxx-cu12==0.48.0
+xlrd==2.0.2
+branca==0.8.2
+chardet==5.2.0
+pycairo==1.29.0
+Authlib==1.6.8
+cuda-core==0.3.2
+sentencepiece==0.2.1
+nvidia-cusparselt-cu12==0.7.1
+matplotlib-venn==1.1.2
+scooby==0.11.0
+fqdn==1.5.1
+gin-config==0.5.0
+ipython-sql==0.5.0
+toml==0.10.2
+PyOpenGL==3.1.10
+weasel==0.4.3
+jsonpointer==3.0.0
+google-auth-httplib2==0.3.0
+spint==1.0.7
+nvtx==0.2.14
+websocket-client==1.9.0
+torchao==0.10.0
+splot==1.1.7
+langgraph-checkpoint==4.0.0
+alabaster==1.0.0
+jaxlib==0.7.2
+google-resumable-media==2.8.0
+namex==0.1.0
+quantecon==0.11.0
+nvidia-cuda-cccl-cu12==12.9.27
+google-cloud-aiplatform==1.138.0
+treelite==4.6.1
+google-cloud-resource-manager==1.16.0
+jupyter_core==5.9.1
+spacy-legacy==3.0.12
+librosa==0.11.0
+ibis-framework==9.5.0
+requests-toolbelt==1.0.0
+smart_open==7.5.1
+tensorflow-metadata==1.17.3
+pysal==25.7
+highspy==1.13.1
+click==8.3.1
+markdown-it-py==4.0.0
+nvidia-cusolver-cu12==11.7.3.90
+cupy-cuda12x==14.0.1
+imutils==0.5.4
+grpclib==0.4.9
+opt_einsum==3.4.0
+folium==0.20.0
+moviepy==1.0.3
+opencv-python==4.13.0.92
+en_core_web_sm==3.8.0
+tensorflow-text==2.19.0
+langchain-core==1.2.15
+yarl==1.22.0
+spacy==3.8.11
+importlib_resources==6.5.2
+peft==0.18.1
+lazy_loader==0.4
+polars-runtime-32==1.35.2
+pylibcudf-cu12==26.2.1
+bigquery-magics==0.10.3
+spanner-graph-notebook==1.1.8
+sqlglot==25.20.2
+linkify-it-py==2.0.3
+types-pytz==2025.2.0.20251108
+tifffile==2026.2.20
+tsfresh==0.21.1
+nbclassic==1.3.3
+scikit-image==0.25.2
+tensorflow_decision_forests==1.12.0
+simsimd==6.5.13
+isoduration==20.11.0
+momepy==0.11.0
+pytest==8.4.2
+nvidia-cuda-nvcc-cu12==12.5.82
+cuda-bindings==12.9.4
+torchsummary==1.5.1
+earthengine-api==1.5.24
+webencodings==0.5.1
+optree==0.19.0
+jax-cuda12-pjrt==0.7.2
+langchain==1.2.10
+safehttpx==0.1.7
+holidays==0.91
+google-cloud-firestore==2.23.0
+fastjsonschema==2.21.2
+pymc==5.28.0
+pydantic==2.12.3
+jaraco.context==6.1.0
+pyogrio==0.12.1
+numba-cuda==0.22.2
+fonttools==4.61.1
+httpimport==1.4.1
+rsa==4.9.1
+tomlkit==0.13.3
+entrypoints==0.4
+anyio==4.12.1
+charset-normalizer==3.4.4
+pooch==1.9.0
+libcuml-cu12==26.2.0
+astropy-iers-data==0.2026.2.23.0.48.33
+ipyleaflet==0.20.0
+cryptography==43.0.3
+missingno==0.5.2
+langgraph==1.0.9
+pandas-datareader==0.10.0
+pyviz_comms==3.0.6
+cycler==0.12.1
+tensorboard==2.19.0
+gast==0.7.0
+jax-cuda12-plugin==0.7.2
+platformdirs==4.9.2
+google-genai==1.64.0
+inflect==7.5.0
+httplib2==0.31.2
+h11==0.16.0
+alembic==1.18.4
+multitasking==0.0.12
+rmm-cu12==26.2.0
+cvxpy==1.6.7
+affine==2.4.0
+cuml-cu12==26.2.0
+pyparsing==3.3.2
+cffi==2.0.0
+h5netcdf==1.8.1
+Markdown==3.10.2
+google-cloud-translate==3.24.0
+rpy2==3.5.17
+regex==2025.11.3
+tf_keras==2.19.0
+google-auth==2.47.0
+nvidia-libnvcomp-cu12==5.1.0.21
+Send2Trash==2.1.0
+cymem==2.0.13
+pylibraft-cu12==26.2.0
+shap==0.50.0
+shapely==2.1.2
+psygnal==0.15.1
+uri-template==1.3.0
+parso==0.8.6
+webcolors==25.10.0
+nltk==3.9.1
+atpublic==5.1
+ImageIO==2.37.2
+sphinxcontrib-applehelp==2.0.0
+bigframes==2.35.0
+pydot==4.0.1
+onemkl-license==2025.3.1
+treescope==0.1.10
+tcmlib==1.4.1
+opentelemetry-sdk==1.38.0
+tiktoken==0.12.0
+nibabel==5.3.3
+multiprocess==0.70.16
+typing_extensions==4.15.0
+PyYAML==6.0.3
+defusedxml==0.7.1
+sphinxcontrib-serializinghtml==2.0.0
+bleach==6.3.0
+tenacity==9.1.4
+python-utils==3.9.1
+google-cloud-bigquery==3.40.1
+google-cloud-bigquery-connection==1.20.0
+opentelemetry-resourcedetector-gcp==1.11.0a0
+ormsgpack==1.12.2
+pydotplus==2.0.2
+pycryptodomex==3.23.0
+openai==2.23.0
+matplotlib==3.10.0
+ml_dtypes==0.5.4
+uvloop==0.22.1
+google-pasta==0.2.0
+giddy==2.3.8
+ipyparallel==8.8.0
+keras==3.10.0
+cuvs-cu12==26.2.0
+mcp==1.26.0
+spacy-loggers==1.0.5
+google-cloud-logging==3.13.0
+rfc3987-syntax==1.1.0
+google-ai-generativelanguage==0.6.15
+keras-hub==0.21.1
+pydata-google-auth==1.9.1
+absl-py==1.4.0
+ydf==0.15.0
+narwhals==2.17.0
+nvidia-cusparse-cu12==12.5.8.93
+openpyxl==3.1.5
+nvidia-cublas-cu12==12.8.4.1
+roman-numerals==4.1.0
+vega-datasets==0.9.0
+mpmath==1.3.0
+etils==1.13.0
+osqp==1.1.1
+traittypes==0.2.3
+opentelemetry-exporter-gcp-monitoring==1.11.0a0
+graphviz==0.21
+google-cloud-trace==1.18.0
+einops==0.8.2
+torchdata==0.11.0
+jax==0.7.2
+cachetools==6.2.6
+aiohappyeyeballs==2.6.1
+annotated-doc==0.0.4
+starlette==0.52.1
+fastapi==0.133.0
+typer==0.24.1
+duckdb==1.3.2
+blinker==1.9.0
+referencing==0.37.0
+googledrivedownloader==1.1.0
+GDAL==3.8.4
+cuda-python==12.9.4
+pycparser==3.0
+et_xmlfile==2.0.0
+jieba==0.42.1
+zict==3.0.0
+hyperopt==0.2.7
+python-louvain==0.16
+SQLAlchemy==2.0.47
+cuda-toolkit==12.8.1
+PyDrive2==1.21.3
+roman-numerals-py==4.1.0
+urllib3==2.5.0
+jaraco.functools==4.4.0
+optax==0.2.7
+pyOpenSSL==24.2.1
+jupyter-console==6.6.3
+libkvikio-cu12==26.2.0
+gspread==6.2.1
+docstring_parser==0.17.0
+albumentations==2.0.8
+jupytext==1.19.1
+seaborn==0.13.2
+librmm-cu12==26.2.0
+cons==0.4.7
+matplotlib-inline==0.2.1
+pynndescent==0.6.0
+stringzilla==4.6.0
+flatbuffers==25.12.19
+omegaconf==2.3.0
+umap-learn==0.5.11
+progressbar2==4.5.0
+pexpect==4.9.0
+torchcodec==0.10.0+cu128
+ptyprocess==0.7.0
+pygame==2.6.1
+kiwisolver==1.4.9
+Cython==3.0.12
+shellingham==1.5.4
+soupsieve==2.8.3
+snowballstemmer==3.0.1
+propcache==0.4.1
+ucxx-cu12==0.48.0
+nbformat==5.10.4
+python-snappy==0.7.3
+rasterstats==0.20.0
+bqplot==0.12.45
+nest-asyncio==1.6.0
+opencv-python-headless==4.13.0.92
+notebook==6.5.7
+flax==0.11.2
+google-cloud-functions==1.22.0
+multipledispatch==1.0.0
+googleapis-common-protos==1.72.0
+eerepr==0.1.2
+torchaudio==2.10.0+cu128
+locket==1.0.0
+prettytable==3.17.0
+pygit2==1.19.1
+fastai==2.8.7
+msgpack==1.1.2
+clarabel==0.11.1
+cligj==0.7.2
+google-cloud-secret-manager==2.26.0
+spglm==1.1.0
+ipytree==0.2.2
+termcolor==3.3.0
+tweepy==4.16.0
+google-cloud-core==2.5.0
+dataproc-spark-connect==1.0.2
+mkl==2025.3.1
+umf==1.0.3
+textblob==0.19.0
+firebase-admin==6.9.0
+simple-parsing==0.1.8
+debugpy==1.8.15
+google-cloud-discoveryengine==0.13.12
+fastcore==1.12.16
+decorator==4.4.2
+pickleshare==0.7.5
+rasterio==1.5.0
+networkx==3.6.1
+typer-slim==0.24.0
+wasabi==1.1.3
+mgwr==2.2.1
+hdbscan==0.8.41
+pydub==0.25.1
+tobler==0.13.0
+more-itertools==10.8.0
+keyrings.google-artifactregistry-auth==1.1.2
+cloudpickle==3.1.2
+nvidia-nvtx-cu12==12.8.90
+fastlite==0.2.4
+colorcet==3.1.0
+lark==1.3.1
+antlr4-python3-runtime==4.9.3
+keras-nlp==0.21.1
+music21==9.9.1
+Pygments==2.19.2
+triton==3.6.0
+toolz==0.12.1
+python-slugify==8.0.4
+sqlparse==0.5.5
+jupyter-leaflet==0.20.0
+gym-notices==0.1.0
+torchvision==0.25.0+cu128
+prophet==1.3.0
+google-cloud-datastore==2.23.0
+semantic-version==2.10.0
+fastprogress==1.1.5
+etuples==0.3.10
+pyspark==4.0.2
+orjson==3.11.7
+terminado==0.18.1
+accelerate==1.12.0
+panel==1.8.7
+apswutils==0.1.2
+pyproj==3.7.2
+sphinxcontrib-htmlhelp==2.1.0
+certifi==2026.1.4
+grpc-interceptor==0.15.4
+pyasn1==0.6.2
+geocoder==1.38.1
+idna==3.11
+mizani==0.13.5
+jupyter_server_terminals==0.5.4
+httpcore==1.0.9
+pyasn1_modules==0.4.2
+ffmpy==1.0.0
+pyperclip==1.11.0
+safetensors==0.7.0
+ndindex==1.10.1
+tblib==3.2.2
+docutils==0.21.2
+scs==3.2.11
+distro==1.9.0
+tf-slim==1.1.0
+babel==2.18.0
+google-cloud-pubsub==2.35.0
+google-api-python-client==2.190.0
+tzlocal==5.3.1
+groovy==0.1.2
+plum-dispatch==2.7.1
+dask==2026.1.1
+blosc2==4.0.0
+sqlalchemy-spanner==1.17.2
+orbax-checkpoint==0.11.33
+wandb==0.25.0
+geopandas==1.1.2
+proglog==0.1.12
+python-dateutil==2.9.0.post0
+tzdata==2025.3
+editdistance==0.8.1
+langsmith==0.7.6
+xarray-einstats==0.10.0
+pydantic_core==2.41.4
+tabulate==0.9.0
+mmh3==5.2.0
+sentry-sdk==2.53.0
+spopt==0.7.0
+dlib==19.24.6
+community==1.0.0b1
+tensorflow==2.19.0
+ale-py==0.11.2
+murmurhash==1.0.15
+notebook_shim==0.2.4
+mdurl==0.1.2
+diffusers==0.36.0
+requests==2.32.4
+Flask==3.1.3
+prometheus_client==0.24.1
+uvicorn==0.41.0
+logical-unification==0.4.7
+soundfile==0.13.1
+itsdangerous==2.2.0
+jsonpatch==1.33
+plotnine==0.14.5
+distributed==2026.1.1
+google-auth-oauthlib==1.2.4
+gdown==5.2.1
+brotli==1.2.0
+py4j==0.10.9.9
+pytensor==2.38.0
+text-unidecode==1.3
+yfinance==0.2.66
+arviz==0.22.0
+cudf-cu12==26.2.1
+wordcloud==1.9.6
+jaraco.classes==3.4.0
+albucore==0.0.24
+python-dotenv==1.2.1
+uritemplate==4.2.0
+nx-cugraph-cu12==26.2.0
+raft-dask-cu12==26.2.0
+hpack==4.1.0
+numexpr==2.14.1
+pydantic-settings==2.13.1
+rapids-logger==0.2.3
+cmake==3.31.10
+pillow==11.3.0
+jsonschema-specifications==2025.9.1
+tables==3.10.2
+google-cloud-storage==3.9.0
+mapclassify==2.10.0
+altair==5.5.0
+filelock==3.24.3
+google-cloud-appengine-logging==1.8.0
+cufflinks==0.17.3
+cvxopt==1.3.2
+six==1.17.0
+watchdog==6.0.0
+sse-starlette==3.2.0
+PySocks==1.7.1
+jupyterlab_widgets==3.0.16
+spaghetti==1.7.6
+intel-cmplr-lib-ur==2025.3.2
+uc-micro-py==1.0.3
+Sphinx==8.2.3
+PyJWT==2.11.0
+google-cloud-bigtable==2.35.0
+numba==0.60.0
+httptools==0.7.1
+rich==13.9.4
+pointpats==2.5.5
+watchfiles==1.1.1
+promise==2.3
+polars==1.35.2
+greenlet==3.3.2
+rfc3986-validator==0.1.1
+threadpoolctl==3.6.0
+opentelemetry-exporter-otlp-proto-http==1.38.0
+libcuvs-cu12==26.2.0
+sniffio==1.3.1
+pylibcugraph-cu12==26.2.0
+holoviews==1.22.1
+pandas-gbq==0.30.0
+frozenlist==1.8.0
+google-crc32c==1.8.0
+torch==2.10.0+cu128
+ipyevents==2.0.4
+libucxx-cu12==0.48.0
+cramjam==2.11.0
+opentelemetry-exporter-otlp-proto-common==1.38.0
+wurlitzer==3.1.1
+confection==0.1.5
+stanio==0.5.1
+easydict==1.13
+argon2-cffi==25.1.0
+llvmlite==0.43.0
+humanize==4.15.0
+rapids-dask-dependency==26.2.0
+argon2-cffi-bindings==25.1.0
+future==1.0.0
+rpds-py==0.30.0
+psycopg2==2.9.11
+iniconfig==2.3.0
+jupyter-events==0.12.0
+nvidia-nccl-cu12==2.27.5
+GitPython==3.1.46
+joblib==1.5.3
+beartype==0.22.9
+hf-xet==1.3.0
+Bottleneck==1.4.2
+apsw==3.51.2.0
+bokeh==3.8.2
+google-cloud-dataproc==5.25.0
+nvidia-cuda-nvrtc-cu12==12.8.93
+colour==0.1.5
+zipp==3.23.0
+blis==1.3.3
+click-plugins==1.1.1.2
+httpx-sse==0.4.3
+nvidia-nvshmem-cu12==3.4.5
+sphinxcontrib-jsmath==1.0.1
+prompt_toolkit==3.0.52
+esda==2.8.1
+param==2.3.2
+google-cloud-speech==2.36.1
+portpicker==1.5.2
+PyWavelets==1.9.0
+google-cloud-monitoring==2.29.1
+Farama-Notifications==0.0.4
+pytz==2025.2
+MarkupSafe==3.0.3
+pyomo==6.10.0
+packaging==26.0
+betterproto==2.0.0b6
+libraft-cu12==26.2.0
+typeguard==4.5.1
+imbalanced-learn==0.14.1
+google-adk==1.25.1
+CacheControl==0.14.4
+ipykernel==6.17.1
+jsonpickle==4.1.1
+xyzservices==2025.11.0
+websockets==15.0.1
+PyGObject==3.48.2
+pandas-stubs==2.2.2.240909
+proto-plus==1.27.1
+segregation==2.5.3
+ratelim==0.1.6
+miniKanren==1.0.5
+geographiclib==2.1
+Jinja2==3.1.6
+frozendict==2.4.7
+libcudf-cu12==26.2.1
+nvidia-cufft-cu12==11.3.3.83
+typing-inspection==0.4.2
+gradio_client==1.14.0
+simplejson==3.20.2
+ruff==0.15.2
+imageio-ffmpeg==0.6.0
+python-json-logger==4.0.0
+cucim-cu12==26.2.0
+jupyter_kernel_gateway==2.5.2
+contourpy==1.3.3
+google-api-core==2.30.0
+opencv-contrib-python==4.13.0.92
+nvidia-cudnn-cu12==9.10.2.21
+opentelemetry-proto==1.38.0
+dask-cudf-cu12==26.2.1
+nvidia-nvimgcodec-cu12==0.7.0.11
+statsmodels==0.14.6
+opentelemetry-exporter-gcp-trace==1.11.0
+deprecation==2.1.0
+tinycss2==1.4.0
+mdit-py-plugins==0.5.0
+tensorflow-datasets==4.9.9
+opentelemetry-api==1.38.0
+langgraph-prebuilt==1.0.8
+keyring==25.7.0
+inequality==1.1.2
+cyipopt==1.5.0
+sympy==1.14.0
+oauth2client==4.1.3
+python-fasthtml==0.12.47
+gspread-dataframe==4.0.0
+wcwidth==0.6.0
+geopy==2.4.1
+natsort==8.4.0
+timm==1.0.25
+rfc3339-validator==0.1.4
+stumpy==1.13.0
+parsy==2.2
+libucx-cu12==1.19.0
+pyerfa==2.0.1.5
+astropy==7.2.0
+curl_cffi==0.14.0
+xarray==2025.12.0
+preshed==3.0.12
+Werkzeug==3.1.6
+SecretStorage==3.5.0
+grpcio==1.78.1
+slicer==0.0.8
+cudf-polars-cu12==26.2.1
+aiosqlite==0.22.1
+grpcio-status==1.71.2
+libpysal==4.14.1
+gitdb==4.0.12
+hyperframe==6.1.0
+opentelemetry-semantic-conventions==0.59b0
+wheel==0.46.3
+h2==4.3.0
+google-cloud-audit-log==0.4.0
+tqdm==4.67.3
+httpx==0.28.1
+cloudpathlib==0.23.0
+thinc==8.3.10
+audioread==3.1.0
+fastdownload==0.0.7
+gcsfs==2025.3.0
+nvidia-nvjitlink-cu12==12.8.93
+access==1.1.10.post3
+tornado==6.5.1
+pandocfilters==1.5.1
+fasttransform==0.0.2
+nvidia-curand-cu12==10.3.9.90
+python-multipart==0.0.22
+yellowbrick==1.5
+jupyter_client==7.4.9
+google-generativeai==0.8.6
+blobfile==3.2.0
+importlib_metadata==8.7.1
+tensorboard-data-server==0.7.2
+attrs==25.4.0
+tbb==2022.3.1
+pluggy==1.6.0
+cuda-pathfinder==1.3.5
+rtree==1.4.1
+arrow==1.4.0
+wrapt==2.1.1
+anywidget==0.9.21
+mlxtend==0.23.4
+smmap==5.0.2
+aiohttp==3.13.3
+opentelemetry-exporter-gcp-logging==1.11.0a0
+sortedcontainers==2.4.0
+pyshp==3.0.3
+sklearn-compat==0.1.5
+xxhash==3.6.0
+zstandard==0.25.0
+Mako==1.3.10
+google-cloud-iam==2.21.0
+autograd==1.8.0
+glob2==0.7
+tensorstore==0.1.81
+tensorflow-probability==0.25.0
+colorlover==0.3.0
+ipyfilechooser==0.6.0
+gradio==5.50.0
+cmdstanpy==1.3.0
+dm-tree==0.1.9
+html5lib==1.1
+python-apt==0.0.0
+PyGObject==3.42.1
+blinker==1.4
+jeepney==0.7.1
+six==1.16.0
+oauthlib==3.2.0
+wadllib==1.3.6
+launchpadlib==1.10.16
+dbus-python==1.2.18
+PyJWT==2.3.0
+importlib-metadata==4.6.4
+httplib2==0.20.2
+zipp==1.0.0
+pyparsing==2.4.7
+lazr.restfulclient==0.14.4
+SecretStorage==3.3.1
+distro==1.7.0
+lazr.uri==1.0.6
+more-itertools==8.10.0
+python-apt==2.4.0+ubuntu4.1
+cryptography==3.4.8
+keyring==23.5.0
+Markdown==3.3.6
+Mako==1.1.3
+MarkupSafe==2.0.1
+packaging==24.1
+inflect==7.3.1
+autocommand==2.2.2
+typeguard==4.3.0
+jaraco.text==3.12.1
+importlib_resources==6.4.0
+wheel==0.43.0
+zipp==3.19.2
+platformdirs==4.2.2
+importlib_metadata==8.0.0
+tomli==2.0.1
+jaraco.collections==5.1.0
+more-itertools==10.3.0
+typing_extensions==4.12.2
+backports.tarfile==1.2.0
+jaraco.functools==4.0.1
+jaraco.context==5.3.0

kaggle_output/wandb/run-20260331_095935-cu87492i/files/wandb-metadata.json

@@ -0,0 +1,42 @@
+{
+  "os":  "Linux-6.6.113+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.12.12",
+  "startedAt":  "2026-03-31T09:59:35.665971Z",
+  "program":  "kaggle.ipynb",
+  "codePath":  "kaggle.ipynb",
+  "email":  "suvraadeep@gmail.com",
+  "root":  "/kaggle/working",
+  "host":  "7a129c53b2b5",
+  "executable":  "/usr/bin/python3",
+  "cpu_count":  2,
+  "cpu_count_logical":  4,
+  "gpu":  "Tesla T4",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "8656922775552",
+      "used":  "7347718606848"
+    }
+  },
+  "memory":  {
+    "total":  "33662472192"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "Tesla T4",
+      "memoryTotal":  "16106127360",
+      "cudaCores":  2560,
+      "architecture":  "Turing",
+      "uuid":  "GPU-c7ebff42-1f98-33a6-9169-a1e3925c4f52"
+    },
+    {
+      "name":  "Tesla T4",
+      "memoryTotal":  "16106127360",
+      "cudaCores":  2560,
+      "architecture":  "Turing",
+      "uuid":  "GPU-4f057d35-8a41-ce55-5d7d-60c77ebb7156"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "8a2fdkg7z7uti6wkzjkth5zruzlcpawp"
+}

kaggle_output/wandb/run-20260331_095935-cu87492i/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_runtime":336,"_wandb":{"runtime":336}}