predict_impulse.py

import joblib
import pandas as pd
import os
from huggingface_hub import hf_hub_download
from opik import Opik, track

# -------------------------------------------------
# Opik API key (HF Spaces / Expo safe)
# -------------------------------------------------
if "OPIK_API_KEY" not in os.environ:
    os.environ["OPIK_API_KEY"] = os.environ.get(
        "EXPO_PUBLIC_OPIK_API_KEY", ""
    )

# -------------------------------------------------
# Load model from HF Model Hub
# -------------------------------------------------
MODEL_REPO = "obx0x3/sensei-model"
MODEL_FILE = "impulse_model.pkl"

model_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename=MODEL_FILE
)

impulse_model = joblib.load(model_path)

# -------------------------------------------------
# Opik client (event logger)
# -------------------------------------------------
try:
    opik_client = Opik(project_name="budgetbuddy-hackathon")
except Exception as e:
    print("Opik disabled:", e)
    opik_client = None


# -------------------------------------------------
# TRACKED FUNCTION (this creates a TRACK)
# -------------------------------------------------
@track(project_name="budgetbuddy-hackathon")
def predict_impulse(category, amount, payment_method, day):
    input_data = {
        "category": category,
        "amount": float(amount),
        "payment_method": payment_method,
        "day": day
    }

    df = pd.DataFrame([input_data])

    pred = impulse_model.predict(df)[0]
    prob = impulse_model.predict_proba(df)[0].max()

    result = {
        "impulsive": bool(pred),
        "confidence": round(float(prob), 3),
        "label": "Impulsive" if pred else "Normal Spend"
    }

    # -------------------------------------------------
    # EVENT inside the TRACK
    # -------------------------------------------------
    if opik_client:
        try:
            opik_client.log_event(
                name="Impulse Analysis Result",
                input=input_data,
                output=result,
                model="sensei-impulse-model",
                metadata={
                    "ui": "hf-space",
                    "feature": "impulse-detection",
                    "confidence_band": (
                        "high" if prob > 0.75 else
                        "medium" if prob > 0.5 else
                        "low"
                    )
                }
            )
        except Exception as e:
            print("Opik logging failed:", e)

    return result