Update app.py

app.py

@@ -72,6 +72,40 @@ def plot_scatter_tab3(subcat, col):
 
     return fig
 
+# Tab 4
+cats = ["Copyright", "Malware", "Unfair/dangerous"] 
+sub_cats = {"Copyright" : ['DRM', 'Encryption', 'Watermarking', 'Patent', 'Trademark',
+ 'Copy left licenses', 'Keygen', 'Reverse engineering',
+ 'Code sharing platforms', 'Public repositories',
+ 'Unauthorized distribution channels'], "Malware" : ['Keylogger', 'Ransomware', 'RAT' ,'Phishing' ,'Rootkit' ,'Worm' ,'Spyware'
+ 'Exploit', 'Adware' ,'Botnet' ,'Anti-detection'], "Unfair/dangerous" : ['Phishing' ,'Biased Code Generation' ,'Cyber Attacks' ,'Model Attacks']}
+
+def rs_change(rs):
+    return gr.Dropdown(choices=list(sub_cats[rs]))
+
+
+def plot_scatter_tab4(cat, subcat, x, y, col):
+    data = raw_data[raw_data["Category"] == cat]
+    data = data[data["Sub-Category"] == subcat]
+    # Group by model and tag
+    grouped_cat = data.groupby(["model", "tag"]).size().reset_index(name="count").sort_values(by="count", ascending=False)
+    grouped_cat["count"] = grouped_cat.groupby(["model"])["count"].transform(lambda x: x / x.sum())
+
+    # Pivot the data for stacking
+    pivot_df = grouped_cat.pivot(index='model', columns='tag', values='count').fillna(0)
+    # pivot_df = pivot_df.sort_values(by="A", ascending=False)
+    # add color vis
+    if col == "Size":
+        pivot_df[col] = pivot_df.index.map(size_map)
+        grouped_cat = grouped_cat.dropna(inplace=True)
+    else:
+        pivot_df[col] = pivot_df.index.str.split("/").str[0]
+
+    # Create an interactive scatter plot
+    fig = px.scatter(pivot_df, x=x, y=y, hover_name=pivot_df.index, title=f'{x} vs {y}', color=col, color_continuous_scale="agsunset")
+
+    # Show the plot
+    return fig
 
 # Gradio Interface
 with gr.Blocks() as demo:
@@ -117,7 +151,7 @@ with gr.Blocks() as demo:
             )
         with gr.TabItem("Helpfulness vs Harmfulness"):
             gr.Markdown("""
-            # Tag vs Tag plot
+            # Helpfulness vs Harmfulness plot
             
             This scatterplot displays for each model the comparison between the rate of Helpful vs Harmful responses.
             You can filter the categories and choose the color of the datapoints based on model or size.
@@ -131,6 +165,15 @@ with gr.Blocks() as demo:
                 ],
                 gr.Plot(label="forecast", format="png"),
             )
+        with gr.TabItem("Category Selection"):
+            category = gr.Radio(choices=list(cats), label="Category Selection")
+            subcategory = gr.Dropdown(choices=[], label="Subcategory Selection")
+            category.change(fn=rs_change, inputs=category, outputs=subcategory)
+            x = gr.Radio(['H', 'A', 'W', 'R'], value="H", label="X-axis Label")
+            y = gr.Radio(['H', 'A', 'W', 'R'], value="R", label="Y-axis Label")
+            col = gr.Radio(['Organisation', 'Size'], value="Organisation", label="Color Label")
+            plot_button = gr.Button("Plot Scatter")
+            plot_button.click(fn=plot_scatter_tab4, inputs=[category, subcategory, x, y, col], outputs=gr.Plot())
 
 # Launch the Gradio app
 demo.launch(share=True)