def expander_text_perplexities()

in utils/gradio_utils.py [0:0]

• 18 lines of code
• 3 McCabe index (conditional complexity)

def expander_text_perplexities(dstats, column_id=""):
    with st.expander(f"Text Perplexities{column_id}", expanded=False):
        st.caption(
            "Use this widget to identify text perplexities from GPT-2."
        )
        st.markdown(
            """
            Outlier perplexities, especially very high values, could highlight 
            an issue with an example. Smaller variations should be interpreted 
            with more care, as they indicate how similar to the GPT-2 training 
            corpus the examples are rather than being reflective of general 
            linguistic properties.
            For more information on GPT-2, 
            see its [model card](https://hf.co/gpt2).
            """
        )
        st.markdown("------")
        st.write(
            "### Here is the list of the examples in the dataset, sorted by "
            "GPT-2 perplexity:"
        )
        if dstats.perplexities_df is None or dstats.perplexities_df.empty:
            st.write(
                "Perplexities have not been computed yet for this dataset, or "
                "this dataset is too large for the UI (> 1,000,000 examples).")
        else:
            st.dataframe(dstats.perplexities_df.reset_index(drop=True))