import pandas as pd
from sklearn.metrics import (
    accuracy_score, precision_score, recall_score,
    f1_score, confusion_matrix
)
import json
import sys


def main():
    if len(sys.argv) < 2:
        print("Usage: python scorer.py <predictions.csv> [truth.csv]")
        sys.exit(1)

    predictions_file = sys.argv[1]
    truth_file = sys.argv[2] if len(sys.argv) > 2 else "data/test.csv"

    try:
        pred = pd.read_csv(predictions_file)
    except FileNotFoundError:
        print(f"Error: predictions file '{predictions_file}' not found.")
        sys.exit(1)

    try:
        truth = pd.read_csv(truth_file)
    except FileNotFoundError:
        print(f"Error: truth file '{truth_file}' not found.")
        sys.exit(1)

    required_pred = {"scenario_id", "prediction"}
    required_truth = {"scenario_id", "label"}

    if not required_pred.issubset(pred.columns):
        print("Error: submission must contain columns: scenario_id, prediction")
        sys.exit(1)

    if not required_truth.issubset(truth.columns):
        print("Error: truth file must contain columns: scenario_id, label")
        sys.exit(1)

    if pred["scenario_id"].duplicated().any():
        dupes = pred.loc[pred["scenario_id"].duplicated(), "scenario_id"].tolist()
        print(f"Error: duplicate scenario_id values in submission: {dupes}")
        sys.exit(1)

    try:
        pred["prediction"] = pd.to_numeric(pred["prediction"])
    except Exception:
        print("Error: prediction column must contain numeric values.")
        sys.exit(1)

    if pred["prediction"].isna().any():
        print("Error: prediction column contains missing values.")
        sys.exit(1)

    non_integer = pred[
        pred["prediction"] != pred["prediction"].astype(int)
    ]

    if len(non_integer) > 0:
        print("Error: prediction column contains non-integer values.")
        print("Predictions must be class labels: 0, 1, or 2.")
        sys.exit(1)

    pred["prediction"] = pred["prediction"].astype(int)

    invalid = set(pred["prediction"].unique()) - {0, 1, 2}
    if invalid:
        print(f"Error: prediction column contains invalid values: {invalid}")
        print("Predictions must be class labels: 0, 1, or 2.")
        sys.exit(1)

    merged = truth.merge(pred, on="scenario_id", how="left", indicator=True)

    missing = merged.loc[merged["_merge"] != "both", "scenario_id"].tolist()
    if missing:
        print(f"Error: missing predictions for {len(missing)} scenario(s): {missing}")
        sys.exit(1)

    y_true = merged["label"]
    y_pred = merged["prediction"]

    results = {
        "accuracy": round(float(accuracy_score(y_true, y_pred)), 4),
        "precision_macro": round(
            float(precision_score(y_true, y_pred, average="macro", zero_division=0)), 4
        ),
        "recall_macro": round(
            float(recall_score(y_true, y_pred, average="macro", zero_division=0)), 4
        ),
        "f1_macro": round(
            float(f1_score(y_true, y_pred, average="macro", zero_division=0)), 4
        ),
        "confusion_matrix": confusion_matrix(y_true, y_pred, labels=[0, 1, 2]).tolist()
    }

    print(json.dumps(results, indent=2))


if __name__ == "__main__":
    main()