Update app.py

app.py

@@ -4,8 +4,6 @@ import torch
 import gradio as gr
 import matplotlib.pyplot as plt
 import numpy as np
-from io import BytesIO
-from PIL import Image
 
 # Load model and tokenizer
 tokenizer = AutoTokenizer.from_pretrained("nlptown/bert-base-multilingual-uncased-sentiment")
@@ -50,32 +48,23 @@ def analyze_text(text):
     # Generate SHAP explanations
     shap_values = explainer([text])
     
-    # Create matplotlib figures for each class
-    images = []
-    for i in range(5):
-        plt.figure(figsize=(10, 3))
-        shap.plots.text(shap_values[:, :, i], display=False)
-        plt.title(f"Feature importance for {output_names_list[i]}")
-        plt.tight_layout()
-        
-        # Save plot to in-memory buffer
-        buf = BytesIO()
-        plt.savefig(buf, format="png", bbox_inches="tight")
-        plt.close()
-        buf.seek(0)
-        
-        # Convert BytesIO to PIL Image
-        image = Image.open(buf)
-        images.append(image)
+    # Create HTML visualizations for all classes
+    html_plots = []
+    for i in range(shap_values.shape[-1]):
+        # Create SHAP text plot and convert to HTML
+        plot_html = shap.plots.text(shap_values[0, :, i], display=False)
+        html_plots.append(plot_html)
     
     # Format confidence scores
     confidence_scores = {model.config.id2label[i]: float(probabilities[i]) 
-                         for i in range(len(probabilities))}
+                        for i in range(len(probabilities))}
     
-    return (predicted_label, confidence_scores, *images)
+    return (predicted_label, 
+            confidence_scores, 
+            *html_plots)
 
-# Create Gradio interface with image components
-with gr.Blocks() as demo:
+# Create Gradio interface with HTML components
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("## 🔍 BERT Sentiment Analysis with SHAP Explanations")
     
     with gr.Row():
@@ -92,7 +81,7 @@ with gr.Blocks() as demo:
         gr.Markdown("""
         ### SHAP Explanations
         Below you can see how each word contributes to different sentiment scores (1-5 stars).  
-        Positive values increase the score, negative values decrease it.
+        Red text increases the score, blue decreases it.
         """)
     
     # Individual Explanation Rows
@@ -100,9 +89,8 @@ with gr.Blocks() as demo:
     for i in range(5):
         with gr.Row():
             plot_components.append(
-                gr.Image(
+                gr.HTML(
                     label=f"Explanation for {model.config.id2label[i]}",
-                    type="pil",
                     elem_classes=f"shap-plot-{i+1}"
                 )
             )
@@ -124,4 +112,4 @@ with gr.Blocks() as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch(share=True)
+    demo.launch(debug = True