"""
predict.py - tiny client for the feature-input fraud model.
Depends only on pandas, numpy, joblib (+ huggingface_hub if you use from_hub).

    model = FraudModel.from_hub("user/repo")      # or FraudModel.load("fraud_model")
    model.predict({"amount": 650, "hour": 3.2, ... all 15 features ...})
"""
import json
import os

import joblib
import pandas as pd

RISK_LABELS = {0: "low", 1: "elevated", 2: "high", 3: "critical"}


class FraudModel:
    def __init__(self, model, config):
        self.model = model
        self.feature_names = config["feature_names"]
        self.thresholds = config["thresholds"]

    @classmethod
    def load(cls, folder="."):
        with open(os.path.join(folder, "config.json")) as f:
            config = json.load(f)
        return cls(joblib.load(os.path.join(folder, "model.joblib")), config)

    @classmethod
    def from_hub(cls, repo_id):
        from huggingface_hub import snapshot_download
        return cls.load(snapshot_download(repo_id))

    def predict(self, features):
        """features: dict | list of dicts | DataFrame  ->  dict (or list of dicts for several rows)"""
        single = isinstance(features, dict)
        df = pd.DataFrame([features] if single else features)
        missing = [c for c in self.feature_names if c not in df.columns]
        if missing:
            raise ValueError(f"Missing features: {missing}")
        X = df[self.feature_names].astype(float)
        if X.isna().any().any():
            raise ValueError("Feature values must not be empty/NaN (use the 'no history' defaults).")
        probs = self.model.predict_proba(X)[:, 1]
        out = []
        for p in probs:
            level = int(sum(p >= t for t in self.thresholds))
            out.append({"fraud_probability": round(float(p), 4), "risk_level": level,
                        "risk_label": RISK_LABELS[level]})
        return out[0] if single else out