Fernanda7171/RandomForestFeatures

1from fastapi import FastAPI, HTTPException2from pydantic import BaseModel3import numpy as np4import pickle5from typing import List6 7app = FastAPI()8 9# Variables globales para el modelo10model = None11threshold = 0.5 12 13def cargar_modelo():14    global model, threshold15    if model is None:16        # Cargamos los archivos que ya tienes en tu Space17        with open("modelo_rf_v4_5feat.pkl", "rb") as f:18            model = pickle.load(f)19        with open("threshold_rf_v4_5feat.pkl", "rb") as f:20            threshold = pickle.load(f)21 22class PredictRequest(BaseModel):23    patient_id: str24    age: float25    activity_level: float26    stress_level: float27    social_support: float28    dia_semana: float29    historial: List[float]30 31@app.get("/health")32def health():33    return {"status": "ok"}34 35@app.post("/predict")36def predict(req: PredictRequest):37    cargar_modelo()38    39    h = np.array(req.historial)40    dia = int(req.dia_semana)41 42    # ── Lógica de Features (idéntica a tu entrenamiento) ──43    if len(h) == 0:44        adh_historica = 0.80; adh_ultimos_3 = 0.80; adh_ultimos_6 = 0.8045        adh_14d = 0.80; estabilidad_14d = 0.0; tendencia_14d = 0.046        ultima_toma = 1; racha_fallos = 0; racha_tomas = 047        max_racha_fallo = 0; n_eventos_previos = 048    else:49        adh_historica = h.mean()50        adh_ultimos_3 = h[-3:].mean() if len(h) >= 3 else h.mean()51        adh_ultimos_6 = h[-6:].mean() if len(h) >= 6 else h.mean()52        ventana_14 = h[-14:]53        adh_14d = ventana_14.mean()54        estabilidad_14d = float(np.std(ventana_14)) if len(ventana_14) > 1 else 0.055        tendencia_14d = adh_14d - adh_historica56        ultima_toma = int(h[-1])57        n_eventos_previos = len(h)58        59        # Cálculo de rachas60        racha_fallos = 061        for t in reversed(h):62            if t == 0: racha_fallos += 163            else: break64        racha_tomas = 065        for t in reversed(h):66            if t == 1: racha_tomas += 167            else: break68        69        max_racha_fallo = cf = 070        for t in h:71            if t == 0:72                cf += 173                max_racha_fallo = max(max_racha_fallo, cf)74            else:75                cf = 076 77    # ── Construcción del Vector (17 features) ──78    features = np.array([[79        req.age, req.activity_level, req.stress_level, req.social_support,80        dia, int(dia in [5, 6]), adh_historica, n_eventos_previos,81        adh_ultimos_3, adh_ultimos_6, ultima_toma, racha_fallos,82        racha_tomas, max_racha_fallo, adh_14d, estabilidad_14d, tendencia_14d83    ]], dtype=np.float32)84 85    # Predicción86    prob = float(model.predict_proba(features)[0][1])87    88    # IMPORTANTE: Convertir tipos de NumPy a nativos de Python para el JSON89    return {90        "patient_id": req.patient_id,91        "prob": round(prob, 4),92        "will_take": bool(prob >= threshold), # Corregido: bool nativo93        "risk": "low" if prob >= threshold else "high",94        "threshold": float(round(threshold, 4)) # Corregido: float nativo95    }