app.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# ----------------------------------
# Initialize FastAPI
# ----------------------------------
app = FastAPI(title="Medical Symptom Prediction API")

# ----------------------------------
# Load Model from Hugging Face Hub
# ----------------------------------
MODEL_NAME = "sm89/Symptom2Disease"

try:
    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
    model.eval()
except Exception as e:
    raise RuntimeError(f"Model loading failed: {e}")

# ----------------------------------
# Label Mapping
# ----------------------------------
id_to_label = {
    0: "Cardiology",
    1: "Dermatology",
    2: "Endocrinology",
    3: "Gastroenterology",
    4: "Infectious",
    5: "Neurology",
    6: "Orthopedics",
    7: "Pulmonology",
    8: "Urology"
}

# ----------------------------------
# Request Schema
# ----------------------------------
class PredictionRequest(BaseModel):
    text: str

# ----------------------------------
# Health Check Endpoint
# ----------------------------------
@app.get("/")
def health_check():
    return {"message": "Medical Symptom API Running"}

# ----------------------------------
# Prediction Endpoint
# ----------------------------------
@app.post("/predict")
def predict(request: PredictionRequest):

    if not request.text.strip():
        raise HTTPException(status_code=400, detail="Text input cannot be empty")

    try:
        inputs = tokenizer(
            request.text,
            return_tensors="pt",
            truncation=True,
            padding=True,
            max_length=128
        )

        with torch.no_grad():
            outputs = model(**inputs)
            probabilities = torch.softmax(outputs.logits, dim=1)

        top_probs, top_indices = torch.topk(probabilities, 3)

        results = []

        for prob, idx in zip(top_probs[0], top_indices[0]):
            label_index = int(idx.item())

            results.append({
                "department": id_to_label.get(label_index, f"LABEL_{label_index}"),
                "confidence": round(float(prob.item()), 4)
            })

        return {
            "input_text": request.text,
            "top_predictions": results,
            "final_prediction": results[0]
        }

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))