api/app.py

from fastapi import FastAPI
from fastapi.responses import JSONResponse
from fastapi.encoders import jsonable_encoder
from pydantic import BaseModel
import numpy as np
import joblib
import torch
from catboost import CatBoostClassifier
from lightgbm import LGBMClassifier
from xgboost import XGBClassifier
from pytorch_tabnet.tab_model import TabNetClassifier

from pathlib import Path
import os
import sys


BASE_DIR = Path(__file__).resolve().parents[1]


if str(BASE_DIR) not in sys.path:
    sys.path.insert(0, str(BASE_DIR))





#  CONFIG 
MODEL_DIR = BASE_DIR/"models"
FUSION_PATH = BASE_DIR/"fusion"/"fusion_model_metadata.joblib"

FEATURES = [
    "rainfall_mm", "humidity_pct", "temp_c", "vegetation_index",
    "elevation_m", "proximity_to_water_km", "wind_speed_mps",
    "surface_water_presence", "daylight_hours", "season_label"
]

#  LOAD MODELS 
print(" Loading models...")

cat_model = CatBoostClassifier()
cat_model.load_model(f"{MODEL_DIR}/catboost.cbm")

xgb_model = joblib.load(f"{MODEL_DIR}/xgboost.joblib")
lgb_model = joblib.load(f"{MODEL_DIR}/lightgbm.joblib")

tabnet_model = TabNetClassifier()
tabnet_model.load_model(f"{MODEL_DIR}/tabnet.zip.zip")
tabnet_scaler = joblib.load(f"{MODEL_DIR}/tabnet_scaler.joblib")

fusion_meta = joblib.load(FUSION_PATH)
print(" All models and fusion metadata loaded successfully.")

# FASTAPI APP 
app = FastAPI(title="Malaria Risk Prediction API", version="1.0")

#  INPUT SCHEMA 
class MalariaInput(BaseModel):
    rainfall_mm: float
    humidity_pct: float
    temp_c: float
    vegetation_index: float
    elevation_m: float
    proximity_to_water_km: float
    wind_speed_mps: float
    surface_water_presence: float
    daylight_hours: float
    season_label: int  # 0 = Dry, 1 = Rainy


#  PREDICTION ENDPOINT 
@app.post("/predict/")
def predict(data: MalariaInput):
    try:
        
        x = np.array([[getattr(data, f) for f in FEATURES]], dtype=float)
        x_scaled = tabnet_scaler.transform(x)

        # Individual model probabilities
        preds = {
            "catboost": float(cat_model.predict_proba(x)[0][1]),
            "xgboost": float(xgb_model.predict_proba(x)[0][1]),
            "lightgbm": float(lgb_model.predict_proba(x)[0][1]),
            "tabnet": float(tabnet_model.predict_proba(torch.tensor(x_scaled).float().numpy())[0][1]),
        }

        # Fusion weights
        models_in_fusion = fusion_meta.get("models", [])
        weights_data = fusion_meta.get("weights", [])
        weights = {m: float(w) for m, w in zip(models_in_fusion, weights_data)}

        # Weighted ensemble score
        risk_score = float(sum(preds[m] * weights[m] for m in preds if m in weights))
        #risk_percentage = round(risk_score * 100, 2)

        # Label decision
        risk_label = "High" if risk_score >= 0.60 else "Medium"

        response = {
            #"input_features": data.dict(),
            "model_outputs": preds,
           # "fusion_weights": weights,
            "risk_score": round(risk_score, 3),
            "risk_label": risk_label
        }

        return JSONResponse(content=jsonable_encoder(response))

    except Exception as e:
        return JSONResponse(
            status_code=500,
            content={"error": str(e), "message": "Prediction failed."}
        )


# ROOT ENDPOINT 
@app.get("/")
def root():
    return {"message": " Malaria Risk Prediction API is running!"}