File size: 6,514 Bytes
8aa6c84
 
 
 
 
 
 
 
 
 
 
 
 
01310fd
 
8aa6c84
 
 
 
 
 
 
 
 
 
 
01310fd
8aa6c84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
"""Regenerate ML artifacts inside Docker to ensure pickle compatibility."""

import numpy as np
import pandas as pd
import json
import joblib
import os
import sys
import warnings

warnings.filterwarnings("ignore")
np.random.seed(42)

_script_dir = os.path.dirname(os.path.abspath(__file__)) if "__file__" in dir() else os.getcwd()
sys.path.insert(0, _script_dir)
from app.services.generators import GENERATORS
from app.services.feature_engine import engineer_features
from app.config import FEATURE_COLS, START_DATE, DAYS

from sklearn.preprocessing import RobustScaler
from sklearn.cluster import KMeans
from sklearn.decomposition import PCA
from sklearn.metrics import silhouette_score
from xgboost import XGBClassifier
import umap

ARTIFACTS_DIR = os.path.join(_script_dir, "app", "artifacts")
os.makedirs(ARTIFACTS_DIR, exist_ok=True)

# 1. Generate dataset
print("Generating dataset...")
counts = {
    "normal_salaried_employee": 600, "normal_freelancer": 350, "normal_student": 450,
    "normal_retiree": 350, "normal_small_business": 300, "normal_high_net_worth": 200,
    "normal_young_professional": 400, "normal_family_household": 350,
    "mule_rapid_passthrough": 130, "mule_structuring_smurfing": 100,
    "mule_funnel_collector": 90, "mule_dormant_burst": 110,
    "mule_recruit_escalation": 120, "mule_round_trip": 100,
    "mule_crypto_cashout": 120, "mule_layering_chain": 110,
    "mule_micro_structuring": 130, "mule_ghost_payroll": 140,
    "mule_onboarding_burst": 120, "mule_device_mule": 110,
}
all_records = []
for btype, count in counts.items():
    print(f"  {btype}: {count}")
    all_records += GENERATORS[btype](count)

df = pd.DataFrame(all_records)
df["timestamp"] = pd.to_datetime(df["timestamp"])
df = df.sort_values("timestamp").reset_index(drop=True)
df["day_of_week"] = df["timestamp"].dt.dayofweek
df["hour"] = df["timestamp"].dt.hour
df["is_weekend"] = df["day_of_week"].isin([5, 6]).astype(int)
df["category"] = df["label"].apply(lambda x: "mule" if x.startswith("mule_") else "normal")
print(f"  Total: {len(df):,} txns, {df['account_id'].nunique():,} accounts")

# 2. Feature engineering
print("Engineering features...")
features_df = engineer_features(df)
label_cat = df.groupby("account_id").agg(label=("label", "first"), category=("category", "first"))
features_df = features_df.join(label_cat)
feature_cols = [c for c in FEATURE_COLS if c in features_df.columns]
X = features_df[feature_cols].fillna(0).values

# 3. Scaler
print("Fitting scaler...")
scaler = RobustScaler()
X_scaled = scaler.fit_transform(X)
X_scaled = np.nan_to_num(X_scaled, nan=0.0, posinf=0.0, neginf=0.0)
joblib.dump(scaler, os.path.join(ARTIFACTS_DIR, "scaler.joblib"))

# 4. PCA
print("Fitting PCA...")
pca2 = PCA(n_components=2)
pca2.fit(X_scaled)
joblib.dump(pca2, os.path.join(ARTIFACTS_DIR, "pca2.joblib"))

# 5. UMAP
print("Fitting UMAP...")
reducer = umap.UMAP(n_components=2, n_neighbors=30, min_dist=0.3, random_state=42)
X_umap = reducer.fit_transform(X_scaled)
joblib.dump(reducer, os.path.join(ARTIFACTS_DIR, "umap_reducer.joblib"))

# 6. KMeans
print("Fitting KMeans...")
K_range = range(2, 16)
sil_scores = []
for k in K_range:
    km = KMeans(n_clusters=k, n_init=10, random_state=42)
    labs = km.fit_predict(X_scaled)
    sil_scores.append(silhouette_score(X_scaled, labs))
best_k = list(K_range)[np.argmax(sil_scores)]
print(f"  Best k = {best_k}")
kmeans = KMeans(n_clusters=best_k, n_init=10, random_state=42)
kmeans.fit(X_scaled)
joblib.dump(kmeans, os.path.join(ARTIFACTS_DIR, "kmeans.joblib"))

# 7. XGBoost classifier
print("Training XGBoost classifier...")
y_binary = (features_df["category"] == "mule").astype(int).values
classifier = XGBClassifier(
    n_estimators=300, max_depth=5, learning_rate=0.1,
    subsample=0.8, colsample_bytree=0.8,
    scale_pos_weight=sum(y_binary == 0) / max(sum(y_binary == 1), 1),
    random_state=42, use_label_encoder=False, eval_metric="logloss",
)
classifier.fit(X, y_binary)
print(f"  Train accuracy: {classifier.score(X, y_binary):.3f}")
joblib.dump(classifier, os.path.join(ARTIFACTS_DIR, "classifier.joblib"))
joblib.dump(classifier, os.path.join(ARTIFACTS_DIR, "surrogate_model.joblib"))

bg_indices = np.random.RandomState(42).choice(len(X), size=min(200, len(X)), replace=False)
np.save(os.path.join(ARTIFACTS_DIR, "shap_background.npy"), X[bg_indices])

# 8. Cluster metadata
print("Computing metadata...")
cluster_labels = kmeans.predict(X_scaled)
features_df["cluster"] = cluster_labels
features_df["umap_1"] = X_umap[:, 0]
features_df["umap_2"] = X_umap[:, 1]

normal_mask = features_df["category"] == "normal"
normal_centroid = X_scaled[normal_mask.values].mean(axis=0).tolist()
normal_distances = np.linalg.norm(X_scaled[normal_mask.values] - np.array(normal_centroid), axis=1)
max_normal_distance = float(np.percentile(normal_distances, 95))

clusters_meta = {}
for c in range(best_k):
    c_mask = features_df["cluster"] == c
    c_data = features_df[c_mask]
    mule_pct = float((c_data["category"] == "mule").mean())
    clusters_meta[str(c)] = {"size": int(c_mask.sum()), "mule_pct": round(mule_pct, 4),
                              "dominant": "mule" if mule_pct > 0.5 else "normal"}

cluster_metadata = {"best_k": best_k, "clusters": clusters_meta,
                    "normal_centroid_scaled": normal_centroid,
                    "max_normal_distance": max_normal_distance, "feature_cols": feature_cols}
with open(os.path.join(ARTIFACTS_DIR, "cluster_metadata.json"), "w") as f:
    json.dump(cluster_metadata, f, indent=2)

# 9. Baseline
normal_features = features_df[normal_mask][feature_cols]
baseline = {"means": normal_features.mean().to_dict(), "stds": normal_features.std().fillna(0).to_dict(),
            "mins": features_df[feature_cols].min().to_dict(), "maxs": features_df[feature_cols].max().to_dict()}
for key in baseline:
    baseline[key] = {k: float(v) for k, v in baseline[key].items()}
with open(os.path.join(ARTIFACTS_DIR, "baseline_features.json"), "w") as f:
    json.dump(baseline, f, indent=2)

# 10. UMAP coordinates
umap_points = [{"x": round(float(row["umap_1"]), 4), "y": round(float(row["umap_2"]), 4),
                "category": row["category"], "label": row["label"]}
               for _, row in features_df.iterrows()]
with open(os.path.join(ARTIFACTS_DIR, "existing_umap_coords.json"), "w") as f:
    json.dump(umap_points, f)

# 11. Transactions CSV
print("Saving CSV...")
df.to_csv(os.path.join(ARTIFACTS_DIR, "synthetic_transactions.csv"), index=False)

print("Done!")