Our current visualization is based on styling pandas tables. This is fine with short sequences where we need to inspect a few items: table visualization does not fit well with longer texts.
I think we should support two HTML-based variants:
- Barplots: (see here for an example)
- Saliency over text: this is standard approach to visualize importance over long sequences. Although we might need to consider recent work that studies how importance is perceived when users are shown such visualization.
Our current visualization is based on styling pandas tables. This is fine with short sequences where we need to inspect a few items: table visualization does not fit well with longer texts.
I think we should support two HTML-based variants: