Moved radar chart to after analysis

app.py

@@ -270,6 +270,37 @@ if selected_x_column != selected_y_column:    # Avoid creating a plot with the s
 else:
     st.write("Please select different columns for the x and y axes.")
 
+
+
+
+# end of custom scatter plots
+st.markdown("## Notable findings and plots")
+
+st.markdown('### Abstract Algebra Performance')
+st.write("Small models showed surprisingly strong performance on the abstract algebra task.  A 6 Billion parameter model is tied for the best performance on this task and there are a number of other small models in the top 10.")
+plot_top_n(filtered_data, 'MMLU_abstract_algebra', 10)
+
+fig = create_plot(filtered_data, 'Parameters', 'MMLU_abstract_algebra')
+st.plotly_chart(fig)
+
+# Moral scenarios plots
+st.markdown("### Moral Scenarios Performance")
+st.write("""
+         While smaller models can perform well at many tasks, the model size threshold for decent performance on moral scenarios is much higher.  
+         There are no models with less than 13 billion parameters with performance much better than random chance. Further investigation into other capabilities that emerge at 13 billion parameters could help
+         identify capabilities that are important for moral reasoning.
+            """)
+
+fig = create_plot(filtered_data, 'Parameters', 'MMLU_moral_scenarios', title="Impact of Parameter Count on Accuracy for Moral Scenarios")
+st.plotly_chart(fig)
+st.write()
+
+
+
+fig = create_plot(filtered_data, 'MMLU_average', 'MMLU_moral_scenarios')
+st.plotly_chart(fig)
+
+
 # Section to select a model and display radar and line charts
 st.header("Compare a Selected Model to the 5 Models Closest in MMLU Average Performance")
 st.write("""
@@ -289,17 +320,15 @@ closest_models_diffs = filtered_data['MMLU_average'].sub(filtered_data.loc[selec
 closest_models = closest_models_diffs.nsmallest(5, keep='first').index.drop_duplicates().tolist()
 
 
-print(closest_models)
-
 # Find the top 10 tasks with the largest differences and convert to a DataFrame
 top_differences_table, top_differences_tasks = find_top_differences_table(filtered_data, selected_model_name, closest_models)
 
 # Display the DataFrame for the closest models and the top differences tasks
 st.dataframe(filtered_data.loc[closest_models, top_differences_tasks])
 
-# Display the table in the Streamlit app
-st.markdown("## Top Differences")
-st.dataframe(top_differences_table)
+# # Display the table in the Streamlit app
+# st.markdown("## Top Differences")
+# st.dataframe(top_differences_table)
 
 # Create a radar chart for the tasks with the largest differences
 fig_radar_top_differences = create_radar_chart_unfilled(filtered_data, closest_models, top_differences_tasks)
@@ -307,34 +336,6 @@ fig_radar_top_differences = create_radar_chart_unfilled(filtered_data, closest_m
 # Display the radar chart
 st.plotly_chart(fig_radar_top_differences)
 
-
-# end of custom scatter plots
-st.markdown("## Notable findings and plots")
-
-st.markdown('### Abstract Algebra Performance')
-st.write("Small models showed surprisingly strong performance on the abstract algebra task.  A 6 Billion parameter model is tied for the best performance on this task and there are a number of other small models in the top 10.")
-plot_top_n(filtered_data, 'MMLU_abstract_algebra', 10)
-
-fig = create_plot(filtered_data, 'Parameters', 'MMLU_abstract_algebra')
-st.plotly_chart(fig)
-
-# Moral scenarios plots
-st.markdown("### Moral Scenarios Performance")
-st.write("""
-         While smaller models can perform well at many tasks, the model size threshold for decent performance on moral scenarios is much higher.  
-         There are no models with less than 13 billion parameters with performance much better than random chance. Further investigation into other capabilities that emerge at 13 billion parameters could help
-         identify capabilities that are important for moral reasoning.
-            """)
-
-fig = create_plot(filtered_data, 'Parameters', 'MMLU_moral_scenarios', title="Impact of Parameter Count on Accuracy for Moral Scenarios")
-st.plotly_chart(fig)
-st.write()
-
-
-
-fig = create_plot(filtered_data, 'MMLU_average', 'MMLU_moral_scenarios')
-st.plotly_chart(fig)
-
 st.markdown("***Thank you to hugging face for running the evaluations and supplying the data as well as the original authors of the evaluations.***")
 
 st.markdown("""