Update app.py

app.py

@@ -1,32 +1,97 @@
 import gradio as gr
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+from sklearn.manifold import TSNE
+from sklearn.cluster import KMeans
+from sentence_transformers import SentenceTransformer
+import io
 from huggingface_hub import InferenceClient
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# Load the dataset
+file_path = 'symbipredict_2022_filtered.csv'  # Ensure this file is uploaded to the Space
+df = pd.read_csv(file_path)
 
+# Load the model from the local directory
+model_path = "all-MiniLM-L6-v2"  # Ensure this directory is uploaded to the Space
+model = SentenceTransformer(model_path)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+# Embed vectors
+embedding_arr = model.encode(df['symptoms'])
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# Apply K-Means with the optimal number of clusters (41 clusters)
+optimal_n_clusters = 41
+kmeans = KMeans(n_clusters=optimal_n_clusters, random_state=42)
+kmeans_labels = kmeans.fit_predict(embedding_arr)
 
-    messages.append({"role": "user", "content": message})
+# Create a DataFrame with prognosis and their corresponding clusters
+cluster_prognosis_mapping = pd.DataFrame({'prognosis': df['prognosis'], 'cluster': kmeans_labels})
+
+# Get the unique cluster-prognosis pairs
+unique_clusters = cluster_prognosis_mapping.drop_duplicates().sort_values(by='cluster')
+
+# Initialize the Inference Client
+client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
-    response = ""
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    query_embedding = model.encode([message])[0]
+    
+    # Combine embeddings with the query embedding
+    combined_embeddings = np.vstack([embedding_arr, query_embedding])
+    
+    # Apply t-SNE to the combined embeddings
+    tsne = TSNE(n_components=2, perplexity=30, n_iter=1000, random_state=42)
+    embedding_tsne = tsne.fit_transform(combined_embeddings)
+    
+    # Separate the transformed query embedding from the rest
+    embedding_tsne_query = embedding_tsne[-1]
+    embedding_tsne = embedding_tsne[:-1]
+    
+    # Plot data along t-SNE components with the query
+    plt.figure(figsize=(14, 10))
+    plt.rcParams.update({'font.size': 16})
+    plt.grid()
+    
+    # Use a colormap for different clusters
+    cmap = plt.get_cmap('tab20', optimal_n_clusters)
+    
+    # Highlight the cluster to which the query embedding belongs
+    query_cluster = kmeans.predict(query_embedding.reshape(1, -1))[0]
+    highlight_cluster = query_cluster
+    
+    c = 0
+    for prognosis in df['prognosis'].unique():
+        idx = np.where(df['prognosis'] == prognosis)
+        if kmeans.predict(embedding_arr[idx])[0] == highlight_cluster:
+            plt.scatter(embedding_tsne[idx, 0], embedding_tsne[idx, 1], c=[cmap(c)] * len(idx[0]), edgecolor='black', linewidth=1, label=f'{prognosis} (Cluster {highlight_cluster})')
+        else:
+            plt.scatter(embedding_tsne[idx, 0], embedding_tsne[idx, 1], c=[cmap(c)] * len(idx[0]), label=prognosis)
+        c = c + 1 / len(df['prognosis'].unique())
+    
+    plt.scatter(embedding_tsne_query[0], embedding_tsne_query[1], c='k', marker='D', s=200, label='query')
+    
+    plt.legend(bbox_to_anchor=(1.05, 1), loc='upper left')
+    plt.xticks(rotation=45)
+    plt.xlabel("t-SNE Component 1")
+    plt.ylabel("t-SNE Component 2")
+    plt.title(f'Query: "{message}" (Belongs to Cluster {highlight_cluster})')
+    
+    # Save the plot to a bytes buffer
+    buf = io.BytesIO()
+    plt.savefig(buf, format='png')
+    buf.seek(0)
+    plt.close()
+    
+    # Generate the text response using the Inference Client
+    messages = [{"role": "system", "content": system_message}]
+    for user_msg, bot_msg in history:
+        if user_msg:
+            messages.append({"role": "user", "content": user_msg})
+        if bot_msg:
+            messages.append({"role": "assistant", "content": bot_msg})
+    messages.append({"role": "user", "content": message})
 
+    response_text = ""
     for message in client.chat_completion(
         messages,
         max_tokens=max_tokens,
@@ -35,13 +100,14 @@ def respond(
         top_p=top_p,
     ):
         token = message.choices[0].delta.content
+        response_text += token
 
-        response += token
-        yield response
+    prognosis_summary = unique_clusters[unique_clusters['cluster'] == highlight_cluster]['prognosis'].tolist()
+    response_text += f"\nThe query belongs to cluster {highlight_cluster} which includes the following prognosis: {', '.join(prognosis_summary)}."
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
+    return {"response": response_text, "image": buf}
+
+# Set up the Gradio Chat Interface
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
@@ -56,8 +122,8 @@ demo = gr.ChatInterface(
             label="Top-p (nucleus sampling)",
         ),
     ],
+    outputs=[gr.Textbox(label="Prognosis Summary"), gr.Image(label="t-SNE Plot")]
 )
 
-
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()