from __future__ import annotations import argparse from pathlib import Path import joblib import pandas as pd from churn_predictor.data import ID_COLUMN, POSITIVE_LABEL, load_inference_data def parse_args() -> argparse.Namespace: parser = argparse.ArgumentParser(description="Run churn predictions on unseen data.") parser.add_argument("--model-path", default="models/best_model.joblib") parser.add_argument("--input-path", default="test.csv") parser.add_argument("--output-path", default="reports/test_predictions.csv") return parser.parse_args() def main() -> None: args = parse_args() model = joblib.load(args.model_path) frame = load_inference_data(args.input_path) if ID_COLUMN not in frame.columns: raise ValueError("Inference data must contain an 'id' column.") feature_frame = frame.drop(columns=[ID_COLUMN]) scores = model.predict_proba(feature_frame)[:, 1] predictions = model.predict(feature_frame) output = pd.DataFrame( { ID_COLUMN: frame[ID_COLUMN], "churn_probability": scores, "predicted_churn_flag": predictions, "predicted_churn_label": [POSITIVE_LABEL if value == 1 else "No" for value in predictions], } ) output_path = Path(args.output_path) output_path.parent.mkdir(parents=True, exist_ok=True) output.to_csv(output_path, index=False) if __name__ == "__main__": main()