Spaces:
Sleeping
Sleeping
File size: 1,905 Bytes
42d30a9 | 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 | import re
class CodeReviewer:
"""
Simulated actor bot that reviews the agent's code edits
for Stage 2 and Stage 3 to detect data leakage and missing scaler.
"""
def review(self, script: str, stage: int) -> str:
script = script or ""
has_scaler = "StandardScaler" in script
# Check if X_scaled = fit_transform(X) happens before train_test_split
# or if scaler is applied to the raw dataframe before splitting
leakage_detected = False
if has_scaler:
if re.search(r"scaler\.fit_transform\(\s*X\s*\)", script):
leakage_detected = True
elif re.search(r"scaler\.fit\(\s*X\s*\)", script):
leakage_detected = True
if stage == 2:
if not has_scaler:
return (
"REJECT: Feature scaling is missing. MLPClassifier is "
"highly sensitive to unscaled features. Add StandardScaler "
"before fitting."
)
if leakage_detected:
return (
"REJECT: You're fitting the scaler on the full dataset. "
"This causes data leakage. Fit only on X_train."
)
return (
"LGTM: Scaler is correctly fitted on training data only. "
"This looks good to merge."
)
if stage == 3:
if leakage_detected or not has_scaler:
return (
"REJECT: Critical data leakage detected. Scaler must be "
"fit on training data only. This invalidates your evaluation."
)
return (
"LGTM: No data leakage detected. Evaluation pipeline "
"looks correct."
)
return "I have no feedback for this stage."
|