File size: 7,143 Bytes
07660e7 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 | """
train.py — Orchestrates training of all three models.
"""
import argparse
import os
import sys
import warnings
import numpy as np
import pandas as pd
from sklearn.metrics import (
accuracy_score,
classification_report,
f1_score,
precision_score,
recall_score,
roc_auc_score,
)
from model import ClassicalMLModel, NaiveBaseline, TraceTransformer
from build_features import get_feature_columns
warnings.filterwarnings("ignore", category=FutureWarning)
def load_data(data_dir: str):
splits = {}
for name in ["train", "val", "test"]:
raw_path = os.path.join(data_dir, f"{name}.parquet")
feat_path = os.path.join(data_dir, f"{name}_features.parquet")
if os.path.exists(raw_path):
splits[f"{name}_raw"] = pd.read_parquet(raw_path)
if os.path.exists(feat_path):
splits[f"{name}_feat"] = pd.read_parquet(feat_path)
return splits
def evaluate_model(y_true, y_pred, y_proba=None, model_name="Model"):
acc = accuracy_score(y_true, y_pred)
prec = precision_score(y_true, y_pred, zero_division=0)
rec = recall_score(y_true, y_pred, zero_division=0)
f1 = f1_score(y_true, y_pred, zero_division=0)
f1_macro = f1_score(y_true, y_pred, average="macro", zero_division=0)
auc = "N/A"
if y_proba is not None and len(np.unique(y_true)) > 1:
try:
auc = f"{roc_auc_score(y_true, y_proba[:, 1]):.4f}"
except Exception:
auc = "N/A"
print(f"\n{'=' * 60}")
print(f" {model_name} — Validation Results")
print(f"{'=' * 60}")
print(f" Accuracy: {acc:.4f}")
print(f" Precision: {prec:.4f}")
print(f" Recall: {rec:.4f}")
print(f" F1 (binary): {f1:.4f}")
print(f" F1 (macro): {f1_macro:.4f}")
print(f" ROC AUC: {auc}")
# safe classification_report: use labels= to avoid crash on single class
all_labels = sorted(set(y_true) | set(y_pred))
names_map = {0: "Normal", 1: "Anomalous"}
target_names = [names_map.get(l, str(l)) for l in all_labels]
print(f"\n{classification_report(y_true, y_pred, labels=all_labels, target_names=target_names, zero_division=0)}")
return {"accuracy": acc, "precision": prec, "recall": rec, "f1": f1, "f1_macro": f1_macro}
def train_naive(splits, model_dir):
print("\n" + "─" * 60)
print(" TRAINING: Naive Baseline (Majority Class)")
print("─" * 60)
train_df = splits.get("train_feat", splits.get("train_raw"))
val_df = splits.get("val_feat", splits.get("val_raw"))
model = NaiveBaseline()
model.fit(None, train_df["label"].values)
y_val = val_df["label"].values
y_pred = model.predict(y_val)
y_proba = model.predict_proba(y_val)
metrics = evaluate_model(y_val, y_pred, y_proba, "Naive Baseline")
save_path = os.path.join(model_dir, "naive_baseline.joblib")
model.save(save_path)
print(f"[SAVED] {save_path}")
return metrics
def train_classical(splits, model_dir):
print("\n" + "─" * 60)
print(" TRAINING: XGBoost (Classical ML)")
print("─" * 60)
train_feat = splits["train_feat"]
val_feat = splits["val_feat"]
feat_cols = get_feature_columns(train_feat)
X_train = train_feat[feat_cols]
y_train = train_feat["label"].values
X_val = val_feat[feat_cols]
y_val = val_feat["label"].values
# check we have both classes
if len(np.unique(y_train)) < 2:
print("[ERROR] Training data has only one class. Cannot train XGBoost.")
print(f" Label distribution: {dict(zip(*np.unique(y_train, return_counts=True)))}")
return {"accuracy": 0, "precision": 0, "recall": 0, "f1": 0, "f1_macro": 0}
model = ClassicalMLModel()
model.fit(X_train, y_train, X_val, y_val)
y_pred = model.predict(X_val)
y_proba = model.predict_proba(X_val)
metrics = evaluate_model(y_val, y_pred, y_proba, "XGBoost")
importance = model.get_feature_importance()
print("\nTop 15 features:")
for feat_name, score in importance.head(15).items():
print(f" {feat_name:40s} {score:.4f}")
save_path = os.path.join(model_dir, "xgboost_model.joblib")
model.save(save_path)
print(f"\n[SAVED] {save_path}")
return metrics
def train_deep(splits, model_dir, num_epochs=3, batch_size=16, lr=2e-5):
print("\n" + "─" * 60)
print(" TRAINING: DistilBERT (Deep Learning)")
print("─" * 60)
train_raw = splits["train_raw"]
val_raw = splits["val_raw"]
X_train = train_raw["raw_trace"].tolist()
y_train = train_raw["label"].tolist()
X_val = val_raw["raw_trace"].tolist()
y_val = val_raw["label"].tolist()
if len(set(y_train)) < 2:
print("[ERROR] Training data has only one class. Cannot train DistilBERT.")
return {"accuracy": 0, "precision": 0, "recall": 0, "f1": 0, "f1_macro": 0}
model = TraceTransformer(
model_name="distilbert-base-uncased",
max_length=512,
batch_size=batch_size,
learning_rate=lr,
num_epochs=num_epochs,
)
model.fit(X_train, y_train, X_val, y_val)
y_pred = model.predict(X_val)
y_proba = model.predict_proba(X_val)
y_val_arr = np.array(y_val)
metrics = evaluate_model(y_val_arr, y_pred, y_proba, "DistilBERT")
save_path = os.path.join(model_dir, "distilbert_trace")
model.save(save_path)
print(f"\n[SAVED] {save_path}")
return metrics
def main():
parser = argparse.ArgumentParser(description="Train anomaly detection models")
parser.add_argument("--data_dir", type=str, default="data/processed")
parser.add_argument("--model_dir", type=str, default="models")
parser.add_argument("--model", type=str, default="all",
choices=["all", "naive", "classical", "deep"])
parser.add_argument("--epochs", type=int, default=3)
parser.add_argument("--batch_size", type=int, default=16)
parser.add_argument("--lr", type=float, default=2e-5)
args = parser.parse_args()
os.makedirs(args.model_dir, exist_ok=True)
splits = load_data(args.data_dir)
results = {}
if args.model in ("all", "naive"):
results["naive"] = train_naive(splits, args.model_dir)
if args.model in ("all", "classical"):
if "train_feat" not in splits:
print("[ERROR] Feature files not found. Run build_features.py first.")
sys.exit(1)
results["classical"] = train_classical(splits, args.model_dir)
if args.model in ("all", "deep"):
if "train_raw" not in splits:
print("[ERROR] Raw data files not found. Run make_dataset.py first.")
sys.exit(1)
results["deep"] = train_deep(splits, args.model_dir,
num_epochs=args.epochs,
batch_size=args.batch_size, lr=args.lr)
if len(results) > 1:
print("\n" + "═" * 60)
print(" MODEL COMPARISON SUMMARY")
print("═" * 60)
summary = pd.DataFrame(results).T
print(summary.to_string())
print("\n[DONE] All training complete.")
if __name__ == "__main__":
main() |