app.py

# app.py
from __future__ import annotations

import warnings
from typing import List, Literal, Optional, Tuple
from config import MODEL_PATH, REAL_LABEL, API_KEY
import joblib
from fastapi import FastAPI, Header, HTTPException
from helper import _combine
from schemas import PredictOut, PredictBatchIn, PredictIn, PredictBatchOut

# Suppress sklearn version warnings
warnings.filterwarnings("ignore", category=UserWarning, module="sklearn")
warnings.filterwarnings("ignore", message=".*InconsistentVersionWarning.*")
# =========================
# Load calibrated model
# (Pipeline: TF-IDF -> CalibratedClassifierCV(LinearSVC))
# =========================
# Additional specific suppression for sklearn version warnings
try:
    from sklearn.exceptions import InconsistentVersionWarning
    warnings.filterwarnings("ignore", category=InconsistentVersionWarning)
except ImportError:
    # Fallback for older sklearn versions
    pass

# Guard against double loading
if 'PIPE' not in globals():
    try:
        print("Loading model from:", MODEL_PATH)
        with warnings.catch_warnings():
            warnings.simplefilter("ignore")
            PIPE = joblib.load(MODEL_PATH)
        print("Model loaded successfully")
    except Exception as e:
        print(f"Error loading model: {e}")
        raise
        
    # Lấy thứ tự class từ estimator cuối để map xác suất cho chắc
    try:
        classes = list(PIPE.named_steps["clf"].classes_)
    except Exception:
        classes = list(getattr(PIPE, "classes_", [0, 1]))  # fallback
        
    print(f"Model classes: {classes}")
    IDX_REAL = classes.index(REAL_LABEL)
    IDX_FAKE = classes.index(0)
    print(f"Real index: {IDX_REAL}, Fake index: {IDX_FAKE}")
else:
    print("Model already loaded, skipping reload...")

# =========================
# Core inference
# =========================
def infer_one(inp: PredictIn) -> PredictOut:
    text_all = inp.text_all.strip().lower() if inp.text_all else _combine(inp.title, inp.text)

    # Suppress warnings during prediction
    with warnings.catch_warnings():
        warnings.simplefilter("ignore")
        probs = PIPE.predict_proba([text_all])[0]
    
    prob_real = float(probs[IDX_REAL])
    prob_fake = float(probs[IDX_FAKE])

    label = "real" if prob_real >= 0.5 else "fake"

    return PredictOut(
        label=label,
        prob_real=prob_real,
        prob_fake=prob_fake,
    )


def infer_batch(items: List[PredictIn]) -> List[PredictOut]:
    return [infer_one(x) for x in items]


# =========================
# FastAPI endpoints
# =========================
app = FastAPI(
    title="SVM Fake/Real News Classifier",
    description="API for classifying news as real or fake using SVM with TF-IDF features",
    version="1.0.0"
)

@app.get("/")
def root():
    return {
        "message": "SVM Fake/Real News Classifier API",
        "endpoints": {
            "predict": "/predict",
            "predict_batch": "/predict_batch",
            "health": "/health"
        },
        "model_info": {
            "classes": ["fake", "real"],
            "model_path": MODEL_PATH,
            "calibrated": True
        }
    }

@app.get("/health")
def health_check():
    return {"status": "healthy", "model_loaded": 'PIPE' in globals()}

@app.post("/predict", response_model=PredictOut)
def predict(payload: PredictIn, x_api_key: str = Header(default="")):
    if x_api_key != API_KEY:
        raise HTTPException(status_code=401, detail="Unauthorized")
    return infer_one(payload)

@app.post("/predict_batch", response_model=PredictBatchOut)
def predict_batch(payload: PredictBatchIn, x_api_key: str = Header(default="")):
    if x_api_key != API_KEY:
        raise HTTPException(status_code=401, detail="Unauthorized")
    return PredictBatchOut(results=infer_batch(payload.items))


if __name__ == "__main__":
    import uvicorn
    print("===== Application Ready =====")
    print("FastAPI app initialized successfully")
    print("API endpoints available at /predict and /predict_batch")
    print("API documentation at /docs")
    print("================================")
    uvicorn.run(app, host="0.0.0.0", port=6778)