import os, json, joblib
import numpy as np
import pandas as pd
from sentence_transformers import SentenceTransformer

def load_bundle(bundle_dir):
    with open(os.path.join(bundle_dir, "config.json"), "r", encoding="utf-8") as f:
        cfg = json.load(f)
    emb_model = SentenceTransformer(cfg["emb_model_name"])
    models = {}
    for c in cfg["label_cols"]:
        p = os.path.join(bundle_dir, f"xgb_{c}.joblib")
        if os.path.exists(p):
            models[c] = joblib.load(p)
    return emb_model, models, cfg

def predict_weighted_risk(texts, emb_model, models, cfg):
    if isinstance(texts, str):
        texts = [texts]
    X = emb_model.encode(list(texts), convert_to_numpy=True)
    label_cols = cfg["label_cols"]
    label_thresholds = cfg["label_thresholds"]
    weights = pd.Series(cfg["weights"])
    risk_thr = cfg["risk_threshold"]

    # Proba
    proba = {}
    for c in label_cols:
        if c in models:
            proba[c] = models[c].predict_proba(X)[:,1]
        else:
            proba[c] = np.zeros(len(texts))
    proba = pd.DataFrame(proba)[label_cols]

    # 0/1 tahmin
    yhat_bin = pd.DataFrame({c: (proba[c] >= label_thresholds.get(c, 0.5)).astype(int) for c in label_cols})

    # Ağırlıklı skor + risk
    weighted = (yhat_bin * weights).sum(axis=1)
    risk_pred = (weighted >= risk_thr).astype(int)

    return {
        "proba": proba,
        "yhat_bin": yhat_bin,
        "weighted_score": weighted,
        "risk_pred": risk_pred
    }

if __name__ == "__main__":
    bundle_dir = "alz_sbert_weighted_export"
    emb, mdl, cfg = load_bundle(bundle_dir)
    sample = "Dün sabah erken kalktım, kahvaltıdan sonra markete gittim. Akşam ailemle buluştum."
    out = predict_weighted_risk(sample, emb, mdl, cfg)
    print("Risk tahmini:", int(out["risk_pred"].values[0]))
    print("Ağırlıklı skor:", float(out["weighted_score"].values[0]))
    print("Etiket 0/1:", out["yhat_bin"].to_dict(orient="list"))