When doing some basic vibe checks for datasets, I realized that the order in which columns are shown might not always be useful for viewing and exploring the data. I might want to quickly show chosen_model next to chosen_response and chosen_avg_rating and continue exploring based on that. Also, this representation might different based on personal preference and search session.
I see it more as an on-the-fly UI thing than a configurable Yaml thing (proposed in #2639), similar to re-ordering a tab in a browser. Based on this, I can also see an option to potentially hide useless columns.
Problem
When doing some basic vibe checks for datasets, I realized that the order in which columns are shown might not always be useful for viewing and exploring the data. I might want to quickly show
chosen_model
next tochosen_response
andchosen_avg_rating
and continue exploring based on that. Also, this representation might different based on personal preference and search session.Related
I noticed an earlier discussion here https://github.com/huggingface/dataset-viewer/issues/2639 (thanks @davanstrien) but it does not fully align with that ideation.
Implementation
I see it more as an on-the-fly UI thing than a configurable Yaml thing (proposed in #2639), similar to re-ordering a tab in a browser. Based on this, I can also see an option to potentially hide useless columns.
Example
Based on UltraFeedback