plots: support multiple Y fields? [qa]

[jorgeorpinel]

Let's say I have a tabular plots file plot.csv:

h1,h2,h3
1,1,0
2,2,0
3,4,-1
4,8,7

and I want to overlay all 3 Hs in a single graph. I can't. I have to pick one at a time, e.g.:

$ dvc plots show -y h2 plot.csv

In fact even with plots diff there's no way to visualize these columns in a single graph (the file would have to be split into 3, each one versioned in a different commit). So what if -y simply accepted multiple field names (or plots diff accepted multiple-y flags)? E.g.:

$ dvc plots show -y h1,h2,h3 plot.csv
# or
$ dvc plots show -y h1 -y h2 -y h3 plot.csv

---

In fact this could also apply to -x. But it could only be done with one or the other, not both (to keep things relatively simple).

[jorgeorpinel]

Cc @dmpetrov

[jorgeorpinel]

Motivation: deep learning or other scenarios where a single stage generates multiple results that need to be evaluated. The user would program the stage to output a single plots file with multiple columns in order to compare many variations. (This is a different option for experiments that doesn't need either run-cache nor multiple commits.)

Of course it won't be very helpful to plot more than 4 or 5 lines in our little Vega graphs, so not 100% sure this has a real application. I just think it's a kind of intuitive feature to add.

[efiop]

Closing for now as stale

[jwgwalton]

+1 on this, i would like to show plots of precision and recall for different population segments and want to have them on the same graph.

[daavoo]

For reference, there is an open discussion #5980 and current WIP on the subject

[daavoo]

+1 on this, i would like to show plots of precision and recall for different population segments and want to have them on the same graph.

Depending on how your plot data is stored, you could use a custom template for that. See iterative/studio-support#23 (comment)

[jwgwalton]

Thanks @daavoo