src/streamlit_app.py

import os
import streamlit as st
from transformers.models.bert import BertTokenizer, BertForSequenceClassification
import torch
import pickle
import random
from collections import defaultdict
import json

# Name encoder yükləmə funksiyası
def load_name_encoder():
    file_path = os.path.join(os.getcwd(), "best_model", "name_encoder.pkl")
    if not os.path.exists(file_path):
        st.error(f"Name encoder faylı tapılmadı: {file_path}")
        st.stop()
    with open(file_path, "rb") as f:
        name_encoder = pickle.load(f)
    return name_encoder

# Model və tokenizer yükləmə
@st.cache_resource
def load_model():
    name_encoder = load_name_encoder()
    model_path = os.path.join(os.getcwd(), "best_model")

    tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
    model = BertForSequenceClassification.from_pretrained(
        model_path,
        num_labels=len(name_encoder.classes_)
    )
    model.eval()
    return tokenizer, model, name_encoder

# Prediction funksiyası
def predict_disease(symptoms_text, tokenizer, model, name_encoder):
    symptoms = [s.strip() for s in symptoms_text.split(",") if s.strip()]
    agg_probs = defaultdict(float)
    n_shuffles = 10

    for _ in range(n_shuffles):
        random.shuffle(symptoms)
        shuffled_text = ", ".join(symptoms)
        inputs = tokenizer(
            shuffled_text,
            return_tensors="pt",
            truncation=True,
            padding=True,
            max_length=128
        )
        with torch.no_grad():
            outputs = model(**inputs)
            probs = torch.nn.functional.softmax(outputs.logits, dim=-1).squeeze()

        for i, p in enumerate(probs):
            agg_probs[i] += p.item()

    for k in agg_probs:
        agg_probs[k] /= n_shuffles

    top_3 = sorted(agg_probs.items(), key=lambda x: x[1], reverse=True)[:3]
    results = []
    for idx, prob in top_3:
        label = name_encoder.classes_[idx]
        results.append({"disease": label, "probability": prob})

    return results

# Page config
st.set_page_config(page_title="Disease API", layout="wide")

# Query parametrlər
query_params = st.query_params
is_api_mode = str(query_params.get("api", ["false"])[0]).lower() == "true"

# Model yüklə
tokenizer, model, name_encoder = load_model()

# API mode
if is_api_mode:
    symptoms = query_params.get("symptoms", [""])[0]
    if symptoms:
        results = predict_disease(symptoms, tokenizer, model, name_encoder)
        api_response = {
            "status": "success",
            "input": symptoms,
            "predictions": results
        }
    else:
        api_response = {
            "status": "error",
            "message": "symptoms parameter required"
        }

    st.markdown(
        f"```json\n{json.dumps(api_response, ensure_ascii=False, indent=2)}\n```"
    )
    st.stop()

# Web interfeys
st.title("🏥 Disease Prediction")
st.success("Model yükləndi!")

# Debug: Siniflər
st.write("Available classes:", list(name_encoder.classes_))

# API usage info
st.markdown("### API İstifadəsi")
space_url = "https://your-username-your-space-name.hf.space"
api_example = f"{space_url}/?api=true&symptoms=fever,cough,headache"
st.code(api_example)

# Form
with st.form(key="predict_form"):
    text = st.text_area("Simptomları daxil edin (vergüllə ayırın):")
    submit_button = st.form_submit_button(label="Predict")

if submit_button:
    if not text.strip():
        st.warning("Simptomları daxil edin!")
    else:
        results = predict_disease(text, tokenizer, model, name_encoder)
        st.subheader("🔍 Nəticələr:")
        for result in results:
            st.write(f"**{result['disease']}** — {result['probability']*100:.2f}%")