from pathlib import Path
import argparse, json
import joblib
import pandas as pd

ROOT = Path(__file__).resolve().parents[1]
MODELS = ROOT / "models"

def load_artifacts():
    metadata = json.loads((MODELS/"model_metadata.json").read_text(encoding="utf-8"))
    model = joblib.load(MODELS/"risk_probability_model.joblib")
    return metadata, model

def predict_cancellation(features: dict) -> dict:
    metadata, model = load_artifacts()
    required = metadata["feature_columns"]
    missing = [c for c in required if c not in features]
    extra = [c for c in features if c not in required]
    if missing:
        raise ValueError(f"Missing required features: {missing}")

    row = pd.DataFrame([{c: features.get(c) for c in required}])
    # The saved risk model is a full sklearn Pipeline:
    # preprocessing + trained classifier. Pass raw model features directly.
    probability = float(model.predict_proba(row)[:, 1][0])
    threshold = float(metadata["risk_threshold"])
    prediction = int(probability >= threshold)

    level = "LOW" if probability < 0.35 else ("MEDIUM" if probability < 0.65 else "HIGH")
    return {
        "target": metadata["target"],
        "cancel_probability": round(probability, 6),
        "threshold": round(threshold, 6),
        "prediction": prediction,
        "prediction_label": "CANCEL_RISK" if prediction else "NOT_CANCEL_RISK",
        "risk_level": level,
        "ignored_extra_fields": extra,
        "model_name": metadata["risk_probability_model"],
        "synthetic_training_data_warning": metadata["synthetic_training_data"],
    }

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("json_file")
    args = parser.parse_args()
    payload = json.loads(Path(args.json_file).read_text(encoding="utf-8"))
    print(json.dumps(
        predict_cancellation(payload.get("features", payload)),
        ensure_ascii=False, indent=2
    ))

if __name__ == "__main__":
    main()