FIX: Fix shape of hist output when input is multidimensional empty list #13368

hershen · 2019-02-05T19:59:10Z

PR Summary

Currently plt.hist([np.array([])]) returns a single array of zeroes for the histogram values (n in the documentation).
There is some pre-processing that converts any input for which np.size(input) == 0 into [np.array([])].

In #13002 the code plt.hist([[], []], color=["k", "r"]) produces an error because the input of [[],[]] is pre-processed into [np.array([])] and its length is no longer equal to the length of color.

The fact that an input of [[],[]] is pre-processed in this way means that its output is a single array of bin values. This seems to contradict the documentation for n:

If input is a sequence of arrays [data1, data2,..], then this is a list of arrays with the values of the histograms for each of the arrays in the same order.

This PR modifies the treatment of multiple empty lists as input to follow the documentation. If the input contains multiple sets of data (even if they're empty), the output will contain the same number of histogram value sets. This also solves #13002.

PR Checklist

Has Pytest style unit tests
Code is Flake 8 compliant
New features are documented, with examples if plot related
Documentation is sphinx and numpydoc compliant
Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way

jklymak · 2019-02-06T23:44:31Z

I'm not 100% following this PR. Why is the new behaviour better than the old? Does it really fix #13002? The test doesn't directly test plt.hist([[],[]], color=['k', 'r']) But more to the point, why are we supporting empty lists to hist?

anntzer · 2019-02-07T08:58:00Z

lib/matplotlib/axes/_axes.py

@@ -6573,7 +6573,7 @@ def hist(self, x, bins=None, range=None, density=None, weights=None,
        # basic input validation
        input_empty = np.size(x) == 0
        # Massage 'x' for processing.
-        if input_empty:
+        if input_empty and len(x) == 0:


Or more simply, this entire if...else block can be deleted and replaced by x = cbook._reshape_2D(x, 'x') which handles empty inputs just fine.

That'll be much cleaner.
But it seems _reshape_2D currently doesn't work with an empty list []:

>>> mpl.cbook._reshape_2D([], 'x') Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/hershen/matplotlib/lib/matplotlib/cbook/__init__.py", line 1418, in _reshape_2D if X.ndim == 1 and not isinstance(X[0], collections.abc.Iterable): IndexError: index 0 is out of bounds for axis 0 with size 0

Is it reasonable to modify it so that it returns [[]] in such a case?

Thanks for noticing that, that's actually a regression due to #11921 that I've re-reported in #13392, which will need to get fixed. I think best would be for this PR to wait for #13392, but I'm not going to hold it up on that.

Well, I noticed it only because of your suggestion ;)
I don't mind waiting for #13392.

It's fixed now.

anntzer · 2019-02-07T09:07:04Z

I think the test is actually fine (it checks that we return a number of bar collections matching the number of inputs).

Supporting empty inputs to hist() is like supporting empty inputs to plot(): "why wouldn't we?"

hershen · 2019-02-07T18:50:10Z

@jklymak, I added context to the PR description.
The new behavior more closely follows the documentation for the output n in hist.
It fixes the issue exposed by #13002 and the code in #13002 does not produce an error message anymore.

It's true that the test doesn't directly test that code. Should I add a test with that exact code?

Currently empty list(s) to hist are supported and produce output (except in cases like #13002). My expectation would be that if output is produced, it's shape will be the same shape as the input (2 sets of input data produce 2 sets of output histogram values, 3 produce 3, etc.). In your opinion, what should happen for inputs of [], [[]], [[],[]]?

jklymak · 2019-02-07T18:57:07Z

Currently empty list(s) to hist are supported and produce output (except in cases like #13002). My expectation would be that if output is produced, it's shape will be the same shape as the input (2 sets of input data produce 2 sets of output histogram values, 3 produce 3, etc.). In your opinion, what should happen for inputs of [], [[]], [[],[]]?

What happens now if you do plt.hist([[], []])? Well, OK, I checked, and its not what this PR proposes:

a, _, _ = plt.hist([[], []]) 
print(a)

yields

array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])

so this is an API change. Is it a good one? I guess so? At the very least needs an API note.

hershen · 2019-02-07T21:54:37Z

Right. With this PR, the output is:

[array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]),
 array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])]

Added an API change entry.

jklymak

Thanks for humouring me @hershen

hershen · 2019-02-08T18:02:56Z

No worries @jklymak!
Sorry for the initial lack of explanations.

doc/api/next_api_changes/2019-02-07-AH.rst

anntzer

anyone can merge post-ci

…ists.

anntzer reviewed Feb 7, 2019

View reviewed changes

hershen force-pushed the empty_hist_with_colors branch from 7b81ca5 to 5c9f024 Compare February 7, 2019 21:50

jklymak approved these changes Feb 7, 2019

View reviewed changes

anntzer reviewed Feb 8, 2019

View reviewed changes

doc/api/next_api_changes/2019-02-07-AH.rst Outdated Show resolved Hide resolved

hershen force-pushed the empty_hist_with_colors branch from c63012e to 7d9857f Compare February 8, 2019 19:20

QuLogic added status: waiting for other PR status: needs revision and removed status: waiting for other PR labels Feb 8, 2019

hershen force-pushed the empty_hist_with_colors branch from 7d9857f to d99ed09 Compare February 11, 2019 17:36

anntzer approved these changes Feb 11, 2019

View reviewed changes

jklymak modified the milestones: v3.0.3, v3.1.0 Feb 11, 2019

jklymak added API: argument checking and removed status: needs revision labels Feb 11, 2019

FIX: Fix shape of hist output when input consists of multiple empty l…

45f29b7

…ists.

hershen force-pushed the empty_hist_with_colors branch from d99ed09 to 45f29b7 Compare February 13, 2019 22:39

timhoffm approved these changes Feb 15, 2019

View reviewed changes

timhoffm merged commit 09b2b0d into matplotlib:master Feb 15, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

FIX: Fix shape of hist output when input is multidimensional empty list #13368

FIX: Fix shape of hist output when input is multidimensional empty list #13368

Uh oh!

hershen commented Feb 5, 2019 •

edited

Loading

Uh oh!

jklymak commented Feb 6, 2019

Uh oh!

anntzer Feb 7, 2019

Uh oh!

hershen Feb 7, 2019

Uh oh!

anntzer Feb 8, 2019

Uh oh!

hershen Feb 8, 2019

Uh oh!

QuLogic Feb 11, 2019

Uh oh!

anntzer commented Feb 7, 2019

Uh oh!

hershen commented Feb 7, 2019

Uh oh!

jklymak commented Feb 7, 2019

Uh oh!

hershen commented Feb 7, 2019

Uh oh!

jklymak left a comment

Uh oh!

hershen commented Feb 8, 2019

Uh oh!

Uh oh!

anntzer left a comment

Uh oh!

Uh oh!

Uh oh!

FIX: Fix shape of hist output when input is multidimensional empty list #13368

FIX: Fix shape of hist output when input is multidimensional empty list #13368

Uh oh!

Conversation

hershen commented Feb 5, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

PR Checklist

Uh oh!

jklymak commented Feb 6, 2019

Uh oh!

anntzer Feb 7, 2019

Choose a reason for hiding this comment

Uh oh!

hershen Feb 7, 2019

Choose a reason for hiding this comment

Uh oh!

anntzer Feb 8, 2019

Choose a reason for hiding this comment

Uh oh!

hershen Feb 8, 2019

Choose a reason for hiding this comment

Uh oh!

QuLogic Feb 11, 2019

Choose a reason for hiding this comment

Uh oh!

anntzer commented Feb 7, 2019

Uh oh!

hershen commented Feb 7, 2019

Uh oh!

jklymak commented Feb 7, 2019

Uh oh!

hershen commented Feb 7, 2019

Uh oh!

jklymak left a comment

Choose a reason for hiding this comment

Uh oh!

hershen commented Feb 8, 2019

Uh oh!

Uh oh!

anntzer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

hershen commented Feb 5, 2019 •

edited

Loading