Tantawi's picture
Upload 13 files
f2a4578 verified
raw
history blame
7.94 kB
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from typing import List
import json
import os
import logging
# Import the existing symptom checker logic
from api_symptom_checker import load_artifacts, predict_symptoms_json
import numpy as np
def safe_predict_symptoms_json(symptoms, model, label_encoder, feature_names):
"""Safe prediction that only uses diseases the label encoder knows about"""
if not symptoms:
return {"error": "No symptoms provided"}
# Build feature vector (convert display names back to feature names)
feature_dict = {name.replace("_", " ").title(): name for name in feature_names}
x = np.zeros(len(feature_names))
matched_symptoms = []
for symptom in symptoms:
if symptom in feature_dict:
feature_name = feature_dict[symptom]
if feature_name in feature_names:
idx = feature_names.index(feature_name)
x[idx] = 1.0
matched_symptoms.append(symptom)
if len(matched_symptoms) == 0:
return {"error": "No valid symptoms found"}
x = x.reshape(1, -1)
# Get predictions - but only use classes the label encoder knows about
proba = model.predict_proba(x)[0]
# SAFETY: Only use the first len(label_encoder.classes_) predictions
max_valid_class = len(label_encoder.classes_)
valid_proba = proba[:max_valid_class] # Only use valid classes
# Get top 3 from valid classes only
top3_idx = np.argsort(valid_proba)[-3:][::-1]
predictions = []
for rank, idx in enumerate(top3_idx, 1):
disease_name = label_encoder.inverse_transform([idx])[0]
confidence = float(valid_proba[idx])
predictions.append({
"rank": rank,
"disease": disease_name,
"confidence": confidence,
"confidence_percent": round(confidence * 100, 2)
})
return {
"input_symptoms": matched_symptoms,
"primary_diagnosis": predictions[0],
"top_predictions": predictions,
"model_confidence": "high" if predictions[0]["confidence"] > 0.7 else "medium" if predictions[0]["confidence"] > 0.4 else "low"
}
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Initialize FastAPI app
app = FastAPI(
title="Symptom Checker API",
description="AI-powered symptom analysis service",
version="1.0.0"
)
# Add CORS middleware
app.add_middleware(
CORSMiddleware,
allow_origins=["*"], # Configure this properly for production
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# Global variables for model artifacts
model = None
label_encoder = None
feature_names = None
# Pydantic models for request/response
class SymptomRequest(BaseModel):
symptoms: List[str]
class PredictionItem(BaseModel):
rank: int
disease: str
confidence: float
confidence_percent: float
class SymptomResponse(BaseModel):
input_symptoms: List[str]
primary_diagnosis: PredictionItem
top_predictions: List[PredictionItem]
model_confidence: str
class AvailableSymptomsResponse(BaseModel):
success: bool = True
symptoms: List[str]
total_symptoms: int
@app.on_event("startup")
async def startup_event():
"""Load model artifacts on startup"""
global model, label_encoder, feature_names
try:
logger.info("Loading symptom checker model artifacts...")
model, label_encoder, feature_names = load_artifacts("symptom_model")
logger.info(f"Model loaded successfully with {len(feature_names)} features")
except Exception as e:
logger.error(f"Failed to load model artifacts: {e}")
raise e
@app.get("/")
async def root():
"""Root endpoint"""
return {
"message": "Symptom Checker API",
"version": "1.0.0",
"endpoints": ["/health", "/api/symptoms", "/api/check-symptoms"]
}
@app.get("/health")
async def health_check():
"""Health check endpoint"""
if model is None:
raise HTTPException(status_code=503, detail="Model not loaded")
return {
"status": "healthy",
"service": "symptom-checker",
"model_loaded": model is not None,
"features_count": len(feature_names) if feature_names else 0
}
@app.get("/api/symptoms", response_model=AvailableSymptomsResponse)
async def get_available_symptoms():
"""Get list of all available symptoms that the model can recognize"""
if feature_names is None:
raise HTTPException(status_code=503, detail="Model not loaded")
# Clean up symptom names for display
clean_symptoms = []
for symptom in feature_names:
# Convert from feature format to readable format
clean_symptom = symptom.replace('_', ' ').title()
clean_symptoms.append(clean_symptom)
return AvailableSymptomsResponse(
success=True,
symptoms=sorted(clean_symptoms),
total_symptoms=len(clean_symptoms)
)
@app.post("/api/check-symptoms")
async def check_symptoms(request: SymptomRequest):
"""Analyze symptoms and return disease predictions"""
global model, label_encoder, feature_names
if model is None or label_encoder is None or feature_names is None:
raise HTTPException(status_code=503, detail="Model not loaded")
if not request.symptoms:
raise HTTPException(status_code=400, detail="No symptoms provided")
try:
# Convert display names back to feature names (Title Case With Spaces -> underscore_format)
feature_symptoms = []
for symptom in request.symptoms:
# Convert "Anxiety And Nervousness" -> "anxiety_and_nervousness"
feature_format = symptom.lower().replace(' ', '_')
feature_symptoms.append(feature_format)
# Use the SAFE prediction logic that handles class mismatch
result = safe_predict_symptoms_json(request.symptoms, model, label_encoder, feature_names)
if "error" in result:
raise HTTPException(status_code=400, detail=result["error"])
# Convert to response format
predictions = []
for pred in result["top_predictions"]:
predictions.append(PredictionItem(
rank=pred["rank"],
disease=pred["disease"],
confidence=pred["confidence"],
confidence_percent=pred["confidence_percent"]
))
# Return format that matches Flutter's SymptomCheckResponse expectations
return {
"success": True,
"predictions": [
{
"rank": pred["rank"],
"disease": pred["disease"],
"confidence": pred["confidence"],
"confidence_percent": f"{pred['confidence_percent']:.2f}%"
}
for pred in result["top_predictions"]
],
"input_symptoms": request.symptoms,
"primary_diagnosis": result["primary_diagnosis"]["disease"],
"model_confidence": result["model_confidence"]
}
except Exception as e:
logger.error(f"Error during symptom prediction: {e}")
raise HTTPException(status_code=500, detail=f"Prediction failed: {str(e)}")
if __name__ == "__main__":
import uvicorn
import os
# Use port 7860 for Hugging Face Spaces, fallback to 8002 for local development
port = int(os.getenv("PORT", 7860))
uvicorn.run("main:app", host="0.0.0.0", port=port, reload=False)