mmagani/lead_scoring_api

1from fastapi import FastAPI, HTTPException2from pydantic import BaseModel3import pickle4import pandas as pd5import numpy as np6import logging7import uvicorn8 9app = FastAPI()10 11# Configure logging12logging.basicConfig(level=logging.INFO)13logger = logging.getLogger(__name__)14 15# Load trained model safely16try:17    model = pickle.load(open("top_tag_model.pkl", "rb"))18    logger.info("✅ Model loaded successfully.")19except FileNotFoundError:20    model = None21    logger.error("❌ Model not found or not loaded.")22 23# Define request model for validation24class PredictionRequest(BaseModel):25    deal_owner: int26    company_name: int27    stage: int28    closed_won_count: int29 30@app.get("/")  # Basic health-check route31async def root():32    return {"message": "API is running"}33 34@app.get("/logs")  # Prevent 404 errors for logs endpoint35async def get_logs():36    return {"message": "Logging data not available"}37 38@app.post("/predict")39async def predict(request: PredictionRequest):40    try:41        logger.info(f"🔄 Received request data: {request.dict()}")42 43        # Convert request data into Pandas DataFrame44        df = pd.DataFrame([request.dict()])45 46        if not model:47            raise HTTPException(status_code=500, detail="Model not found or not loaded")48 49        # Get probabilities for all tags50        probabilities = model.predict_proba(df)51 52        # Extract top 5 tags with highest probabilities53        top_5_indices = np.argsort(probabilities[0])[-5:][::-1]  # Sort in descending order54        top_5_tags = [{"tag_index": int(idx), "probability": float(probabilities[0][idx])} for idx in top_5_indices]55 56        logger.info(f"✅ Top 5 tags predicted: {top_5_tags}")57        return {"top_5_tags": top_5_tags}58 59    except Exception as e:60        logger.error(f"❌ Prediction error: {e}")61        raise HTTPException(status_code=500, detail=str(e))62 63# Run FastAPI server when executed directly64if __name__ == "__main__":65    uvicorn.run(app, host="0.0.0.0", port=7860)