regression: bar plot with multi-column category doesn't work anymore

[soxofaan]

Code Sample, a copy-pastable example if possible

Doing a bar plot using a multi-column category used to work

df.plot.barh(x=['fruit', 'animal'], y='size')

In pandas 0.22.0:

since pandas 0.23.0:

Is it intended that this stopped working?
I kind of liked the feature.

[soxofaan]

using some git bisect magic I found that this commit introduces the ValueError:
d7d8f2d, which refers to PR #18695 and issue #18671

[soxofaan]

also related: #19699

[gfyoung]

cc @TomAugspurger : Labeled as regression, but of course, feel free to modify if change was deliberate.

[daminisatya]

Is the issue still open? If so, I would like to work on this.

[TomAugspurger]

cc @masongallo if you recall the discussion around this.

[soxofaan]

@daminisatya yes, as far as I know this is not yet resolved

[masongallo]

A few comments:

• The change was intentional to conform with the api & docs
• The signature of barh in pandas is quite confusing to me: df.plot.barh(x=['fruit', 'animal'], y='size') but the fruit,animal actually appears on the y-axis and size appears on the x-axis!
• The signature of ax.barh in matplotlib actually takes the y-value as its first arg

[soxofaan]

@masongallo :
I just used barh to make sure the screenshot was easily readable,
The "feature" also used to work for standard bar in pandas 0.22:

The change was intentional to conform with the api & docs

I understand for the use case of line and scatter plots,
but for categorical plots like bar (and pie see lower), I think it would be an interesting feature to bring back support for multi-column based labels. Is there interest for that @TomAugspurger ?

The signature of barh in pandas is quite confusing to me

In my mind, the x/y handling in barh makes sense as I see barh as a flipped version of bar: the x-axis is drawn vertically and y-axis is drawn horizontally. During data exploration, I usually switch between bar and barh functions names to improve readability of the labels, and in that process I don't want to also change the arguments of those functions. That being said, I think the axis handling of barh is probably another discussion and out of scope of this ticket.

For completeness I also tried it out with pie, but there the x argument seems to be ignored completely, so it kind of works with both 0.22 and 0.23. However, the pie piece labels come from the index instead of the given x column. But that is probably another (known) issue.

[TomAugspurger]

Allowing multiple `x` might make sense for bar / barh. FWIW, I would personally do this with df.set_index(['fruit', 'animal'])['size'].plot.bar()

…

On Wed, Jun 13, 2018 at 4:12 AM, Stefaan Lippens ***@***.***> wrote: @masongallo <https://github.com/MasonGallo> : I just used barh to make sure the screenshot was easily readable, The "feature" also used to work for standard bar in pandas 0.22: [image: screen shot 2018-06-13 at 10 48 04] <https://user-images.githubusercontent.com/44946/41340291-521128e2-6ef7-11e8-821e-be14f586f56e.png> The change was intentional to conform with the api & docs I understand for the use case of line and scatter plots, but for categorical plots like bar (and pie see lower), I think it would be an interesting feature to bring back support for multi-column based labels. Is there interest for that @TomAugspurger <https://github.com/TomAugspurger> ? The signature of barh in pandas is quite confusing to me In my mind, the x/y handling in barh makes sense as I see barh as a flipped version of bar: the x-axis is drawn vertically and y-axis is drawn horizontally. During data exploration, I usually switch between bar and barh functions names to improve readability of the labels, and in that process I don't want to also change the arguments of those functions. That being said, I think the axis handling of barh is probably another discussion and out of scope of this ticket. For completeness I also tried it out with pie, but there the x argument seems to be ignored completely, so it kind of works with both 0.22 and 0.23. However, the pie piece labels come from the index instead of the given x column. But that is probably another (known) issue. [image: screen shot 2018-06-13 at 10 49 45] <https://user-images.githubusercontent.com/44946/41340518-e0ed1c4c-6ef7-11e8-9b93-ed97ea6b3bca.png> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#21386 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHInXYqoecXRFXxguCDPJM7pyzrFWgks5t8NdogaJpZM4UgdDm> .