DOC Rework GapEncoder example

[LilianBoulard]

Extracted from #546.

[jovan-stojanovic]

Thanks @LilianBoulard, this looks very good! A few comments.

[GaelVaroquaux]

LGTM, but I think that we also need to change the index because there is a link to the example which has been renamed, no?

[jovan-stojanovic]

You should adapt the example to the fact that #581 was merged

[jovan-stojanovic]

Thanks @LilianBoulard! As agreed, I will take over this PR so we move faster.

[Vincent-Maladiere]

Great concision effort again (-100!). My main remark regards using "dirty" for variable names and explainers. This can be easily misinterpreted, and I'd rather have "high cardinality, "high entropy" or "noisy" instead. WDYT?

[Vincent-Maladiere]

This part (and the subsequent visualization) is great! A real wow effect 👍

[jovan-stojanovic]

My main remark regards using "dirty" for variable names and explainers. This can be easily misinterpreted, and I'd rather have "high cardinality, "high entropy" or "noisy" instead. WDYT?

You have a point, looking back at the example I realize that we introduce "dirty" without explaining what it means. I will correct this.

This term "dirty" is something that was used in the literature and in dirty-cat to describe data whose source of variation can be anything, meaningful and meaningful as opposed to "noisy" for instance, which has different connotation (often synonym to meaningless) in some fields (e.g. econometrics, signal processing).

As for alternatives, I like "high cardinality" (also used in literature), "high entropy" may be less known among our users.

[jovan-stojanovic]

WDYT of this version? @Vincent-Maladiere

[Vincent-Maladiere]

This term "dirty" is something that was used in the literature and in dirty-cat to describe data whose source of variation can be anything

Yes, we used it in a paper, but it doesn't mean it conveys the right message 😅. External users might be surprised by this wording, like Cailean Osborne —he was in high spirits when he found out we call those "dirty"!

WDYT of this version? @Vincent-Maladiere

Well, I still prefer using "high cardinality" if you like. Let's bring this up during the next meeting.

[Vincent-Maladiere]

LGTM!