app.py

import gradio as gr
import numpy as np
import pandas as pd
from sklearn.cluster import MiniBatchKMeans
from sentence_transformers import SentenceTransformer
from functools import lru_cache
from huggingface_hub import AsyncInferenceClient
from sklearn.decomposition import PCA

# ------ Data Loading ------
df = pd.read_csv("symbipredict_2022_filtered.csv")

# ------ Model Initialization ------
model = SentenceTransformer("all-MiniLM-L6-v2")
embedding_arr = model.encode(df['symptoms']).astype(np.float32)

# ------ Clustering Setup ------
kmeans = MiniBatchKMeans(n_clusters=10, random_state=42)
cluster_labels = kmeans.fit_predict(embedding_arr)
cluster_prognosis_map = df.groupby(cluster_labels)['prognosis'].agg(
    lambda x: x.value_counts(normalize=True).head(3).to_dict()
)

# ------ PCA Initialization ------
pca = PCA(n_components=2).fit(embedding_arr)

# ------ Cached Functions ------
@lru_cache(maxsize=100)
def cached_encode(text):
    return model.encode(text, convert_to_numpy=True)

# ------ Async Inference Client ------
client = AsyncInferenceClient("HuggingFaceH4/zephyr-7b-beta")

# ------ Streaming Response Function ------
async def respond(message, history, system_message):
    try:
        # Encoding and clustering
        query_embedding = cached_encode(message)
        query_cluster = kmeans.predict(query_embedding.reshape(1, -1))[0]
        
        # Generate streaming response
        stream = await client.chat_completion(
            messages=[{
                "role": "system",
                "content": system_message
            }, {
                "role": "user",
                "content": message
            }],
            max_tokens=512,
            stream=True,
            temperature=0.7,
            top_p=0.95
        )
        
        full_response = ""
        async for chunk in stream:
            content = chunk.choices[0].delta.content
            if content:
                full_response += content
        
        # Format prognosis likelihoods
        prognosis_info = cluster_prognosis_map[query_cluster]
        formatted_prognoses = ", ".join([f"{k}: {v*100:.1f}%" for k, v in prognosis_info.items()])
        
        # Append formatted prognosis to response
        yield f"{full_response}\n\nCommon prognoses: {formatted_prognoses}"
        
    except Exception as e:
        yield f"Error: {str(e)}"

# ------ Gradio Interface ------
with gr.Blocks() as demo:
    gr.Markdown("# Medical Diagnosis Assistant")
    chatbot = gr.ChatInterface(
        respond,
        additional_inputs=[
            gr.Textbox(value="Medical diagnosis assistant", label="System Role")
        ],
        examples=[["I have a headache and fever"]]  # Example query for guidance
    )

if __name__ == "__main__":
    demo.launch(max_threads=10)