[FEAT] Add a plot_topics method to GapEncoder

[Vincent-Maladiere]

Problem Description

One of the GapEncoder 's main features is its interpretability. The documentation shows how to plot a heatmap representing activations across topics and their content, but getting this representation right is time-consuming.

Feature Description

I suggest we bring this method directly into the GapEncoder class.

Alternative Solutions

Creating a separate function elsewhere, akin to sklearn.tree.plot_tree, but at the cost of being less practical and visible.

Additional Context

No response

[jeromedockes]

I agree this would be useful; see also #909 (comment)

[GaelVaroquaux]

Yup! But can we have a generic methods (ie get_report()), which could be used everywhere in skrub?

[Vincent-Maladiere]

But can we have a generic methods (ie get_report()), which could be used everywhere in skrub?

Can you elaborate a bit on this @GaelVaroquaux? Do you see get_report also working for TextEncoder and MinHashEncoder?

[GaelVaroquaux]

Can you elaborate a bit on this @GaelVaroquaux? Do you see get_report also working for TextEncoder and MinHashEncoder?

Maybe. But to me, it's more a question of something homogeneous in the library: we could have this on the Recipe (or whatever, I don't know). I would love it if people would have a one-stop command to ask an object: give me a useful visualization of what you have in your belly. Something a bit more powerful than the default repr.

[Vincent-Maladiere]

Very interesting! Could we list somewhere the kind of visualizations we would like to see for some objects?

[jeromedockes]

Very interesting! Could we list somewhere the kind of visualizations we would like to see for some objects?

yes let's make a list and do a sprint :)