-
-
Notifications
You must be signed in to change notification settings - Fork 7.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add support for multiple hatches, edgecolors and linewidths in histograms #28073
base: main
Are you sure you want to change the base?
Conversation
I'd suggest showing what this does with an example either in the GitHub pr description, or ideally in the gallery |
I'm not really sure if I need to add a new test or just modify an exisiting one(test_hist_stacked_bar) in test_axes.py |
I guess there are two things here:
|
pinging @story645 for review. The failing tests are unrelated |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
also needs updated documentation that the patch properties are now vectorized (@timhoffm any concerns here?)
What should we be concerned about? We already have vectorized
This should also get a what's new entry. |
My bias is vectorize everything so I don't have concerns, but in the past for some vectorization discussions there have been concerns about the tradeoffs. But if there isn't opposition, awesome! |
I don't see any drawbacks for |
So we are planning to vectorize all parameters of Patches? Like joinstyle, capstyle etc. |
Not at this time w/ the current architecture, especially because nobody has asked for those. |
150c63b
to
2351cbd
Compare
Codecov is acting fishy, it passed once and failed again after squashing. Anything else to add/change? |
As of the last commit, this behavior is consistent for all histtypes unless mentioned otherwise:
|
@Impaler343 I started trying to flow chart that and then realized it's probably clearer as a table. Let me know if this jives with what you're saying:
|
Have bolded the corrected ones |
I'm unable to fix CircleCI errors for docs. Could someone help me out? |
Hi, so the error is in your what's new: /home/circleci/project/doc/users/next_whats_new/histogram_vectorized_parameters.rst:47: WARNING: Explicit markup ends without a blank line; unexpected unindent. error: https://app.circleci.com/pipelines/github/matplotlib/matplotlib/31532/workflows/2533c013-7833-4d0c-ae72-dead7c7fbc76/jobs/83481?invite=true#step-113-207133_109 |
Ok it turns out the vectorization of
|
I think this input is ambiguous cause it can mean either:
So I think it's ok to error out (unless/until someone comes around expecting 1 or 2) but we should maybe special case and have a better error message, like if not all(colors) (if any are NaN) "Ambiguous color specification: colors in the list may not be None" Or something like that. |
The added lines have all been covered, codecov seems to still fail |
Can parameterize this as a kwargs arg dict |
Oh yes! I completely missed this method |
Updated table by expanding the first few cases and correcting:
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
please don't special case none. Here you end up only checking the alpha channel. Otherwise, I think this hits the balance of everything tested but way fewer images tests?
PR cleanliness test is failing on the added/removed test images, that can get fixed by a squash rebase |
Specify extensions for test Added modified baseline images Modified test for histogram with single parameters Fixed test Add modified baseline images Made changes concise according to suggestion Made a more detailed gallery example Fix Docs Added whats new note, documentation for vectorization, doc fix Added new test, and reverted changes in old test Added baseline images Modified test to pass codecov, added plot in whats new entry Fix test Added baseline images Altered whats new entry, docs and gallery example Resolved edgecolor and facecolor setting Minor fix Fix docs Modified files to include facecolor and added test Removed figsize from test Add multiple baseline image names Fixed test? Fixed test? Removed parametrize usage Add baseline images Add baseline image Fix docs Fix docs Deleted baseline images, changed test Fix test Fix test Handled None array passing to color Handled passing None list Modified nested patch condition Minor Fix Grammar nits Modified test, edited None handling in sequence so that it errors out
Separated tests Modified test Modified test to pass by using halved zorders
Edited expected facecolors for step Changed blue to C0
lib/matplotlib/tests/test_axes.py
Outdated
@@ -4614,6 +4672,10 @@ def test_hist_emptydata(): | |||
ax.hist([[], range(10), range(10)], histtype="step") | |||
|
|||
|
|||
def test_hist_none_patch(): | |||
plt.hist([1, 2], label=["First", "Second"]) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
plt.hist([1, 2], label=["First", "Second"]) | |
plt.hist([ ], label=["First", "Second"]) |
None patch is the empty list/array case
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
And what you probably want to test is
assert len(patches) == 0
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This too works, basically the None condition is hit when we don't have enough datasets to cover each label. This is the doing of itertools.zip_longest
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should i still change it up? As the previous one seems clearer
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry, I thought that zip_longest would copy the patch to both labels. I think there needs to be some sort of assert that you're test is doing what you think it is - maybe then checking that len(patches) != len(labels)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What is len(lbs) here?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The number of labels that have been used in the plot
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry, I meant what were the actual numbers/does this test pass.
But yes if this test passes and coverage improves (add a comment that this test is to cover the if not patch case), then I'll let this concern go.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
But also assert len(patches) < len(labels) too I think?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes the tests pass, and yes we can add the length(patches) < len(labels) one as well, does the same comparison.
The actual numbers are 1 < 2
Cause I think this maybe got lost, I was suggesting that this construction makes the skip case clearer: for patch, lbl in itertools.zip_longest(patches, labels):
if not patch: continue
p = patch[0]
kwargs.update({
'hatch': next(hatches),
'linewidth': next(linewidths),
'linestyle': next(linestyles),
'edgecolor': next(edgecolors),
'facecolor': next(facecolors),
})
p._internal_update(kwargs)
if lbl is not None:
p.set_label(lbl)
for p in patch[1:]:
p._internal_update(kwargs)
p.set_label('_nolegend_') |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for all your patience and extra thanks for tracing through and figuring out and testing our histogram color semantics/precedence 😅
I think the plot tests are now at a state where it's easy to see which parameter failed but there's also now only six image tests.
I'm pretty sure that the color semantics test doesn't try to draw anything but there might be a "build only artist tree" function that I'm missing.
lib/matplotlib/tests/test_axes.py
Outdated
def test_hist_none_patch(): | ||
# To cover None patches when excess labels are provided | ||
labels = ["First", "Second"] | ||
patches = [[1, 2]] | ||
fig, ax = plt.subplots() | ||
ax.hist(patches, label=labels) | ||
_, lbls = ax.get_legend_handles_labels() | ||
assert (len(lbls) < len(labels) and len(patches) < len(labels)) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you please elaborate what this tests? What are None patches? Does this express what you want to test?
def test_hist_none_patch(): | |
# To cover None patches when excess labels are provided | |
labels = ["First", "Second"] | |
patches = [[1, 2]] | |
fig, ax = plt.subplots() | |
ax.hist(patches, label=labels) | |
_, lbls = ax.get_legend_handles_labels() | |
assert (len(lbls) < len(labels) and len(patches) < len(labels)) | |
def test_hist_unused_labels(): | |
# When a list with one dataset and N elements is provided and N labels, ensure | |
# hat the first label is used for the dataset and all other labels are ignored | |
fig, ax = plt.subplots() | |
ax.hist([[1, 2, 3]], label=["values", "unused", "also unused"]) | |
_, labels = ax.get_legend_handles_labels() | |
assert labels == ["values"] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
None patches are the ones generated when the number of labels specified are more than the number of patches(Filled in by zip_longest). The suggested test also covers the line and has clearer labels, ill change it up
lib/matplotlib/tests/test_axes.py
Outdated
for i, (xi, value) in enumerate(zip(x, values)): | ||
axr.hist(xi, bins=bins, histtype=histtype, **{kw: value}, | ||
zorder=(len(x)-i)/2) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I believe the index/zorder acrobatics is only to get the draw order. It's simpler to do this via
for i, (xi, value) in enumerate(zip(x, values)): | |
axr.hist(xi, bins=bins, histtype=histtype, **{kw: value}, | |
zorder=(len(x)-i)/2) | |
for x, value in reversed(zip(xs, values)): | |
axr.hist(x, bins=bins, histtype=histtype, **{kw: value}) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
did you fix the typo:
for xi, value in reversed(zip(x, values)):
axr.hist(xi, bins=bins, histtype=histtype, **{kw: value})
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
for x, value in reversed(list(zip(xs, values))):
axr.hist(x, bins=bins, histtype=histtype, **{kw: value})
yes, I tried this one
# * linestyles | ||
# | ||
# | ||
# Edge-Colors |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not convinced that there is much to learn by having individual examples for every parameter. This feels lengthy without providing much additional insight. Maybe the whats new example or 2 individually crafted examples (one simple and one using multiple values simultaneously (e.g. edgecolor, facecolor, hatch) are better?
But I won't block over this. However, if you keep this, please use a consistent section naming, preferably as the parameter is written.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
what's most important to me is that the example only set one parameter rather than trying to set multiple. But also the reason I encouraged this approach was b/c https://matplotlib.org/devdocs/gallery/pie_and_polar_charts/pie_features.html#sphx-glr-gallery-pie-and-polar-charts-pie-features-py was in response to folks not knowing that they could set all these parameters. Basically, I wanted something that would pop up in a search.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
My point is that these styling parameters all work conceptually similar. Repeating the same pattern for every parameter feels a bit lengthy and boring to me. I think the pie parameters are somewhat different in that they have different effects and one cannot guess what one is doing based on the example of another.
But it's not worth a longer discussion. If you feel they are all valuable, I won't block over that.
#28533 reminded me that this should get a versionadded directive https://matplotlib.org/devdocs/devel/api_changes.html#announce-new-and-deprecated-api |
datasets in *x*: | ||
*edgecolors*, *facecolors*, *linewidths*, *linestyles*, *hatches*. | ||
|
||
.. versionadded:: 3.10 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry wgough
.. versionadded:: 3.10 | |
.. versionadded:: 3.10 | |
List input to hatch parameters |
Sorry forgot to add, here especially you should briefly describe what's added. My sentence here is super clunky English, so more here for example than something you should stick with.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I didnt get you....
PR summary
Closes #26718 Distributes keyword args passed to each Patch using a cycler. Probably not the best way to do this?
PR checklist