Add subsampling hint in TableReport used in expression

[GaelVaroquaux]

At the bottom of the first tab of the TableReport, there is some space (where it says "100 rows ✕ 9 columns. "):

We should use this to add usability messages in general, when the TableReport is called from a given context.

Specifically, in the case of subsampled previews, we should write here "100 rows ✕ 9 columns (subsampled from xxxx rows)".

[Vincent-Maladiere]

Should we remove this line from the display? It is kinda redundant with your suggestion, but since the display might be switched to a regular pandas or polars dataframe in the future as per #1377, we might have to keep it.

[GaelVaroquaux]

For the reason that you give, I think that for now we should not remove the line on the top.

I do think that we can be more creative with the space on the top, and have a better visual display of the information. But I still think that the information that this is working on a subsample should be very very visible, and hence duplication is fine

[Vincent-Maladiere]

when the TableReport is called from a given context.

In which context should we provide the unsampled number of rows? As @rcap107 pointed out, if we simply evaluate the expression using the full data to get the number of samples of the output, then we don't benefit from subsampling at all.

[Vincent-Maladiere]

An alternative is to simply mention "subsampled", without the number of rows

[GaelVaroquaux]

An alternative is to simply mention "subsampled", without the number of rows

Yes, that's the right way to do it IHMO (I wrote the original issue a bit too fast, without thinking it through). And the corresponding API in the TableReport could be something like footer_msg="(subsampled)" (well, "" by default, but called here with "(subsampled)".

Thanks!!

[Vincent-Maladiere]

Done in Add subsampling hint in TableReport used in expression #1418 :)