Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix #3886) #4119

umairidris · 2015-02-17T23:28:06Z

efiring · 2015-02-17T23:55:18Z

lib/matplotlib/tests/test_axes.py

@@ -482,6 +482,9 @@ def test_hexbin_extent():

    ax.hexbin(x, y, extent=[.1, .3, .6, .7])

+def test_hexbin_empty():
+    # From #3886: creating hexbin from empty dataset raises ValueError
+    ax.hexbin([], [])


You need to create the Axes first. Also, I think you will want to use the @cleanup decorator.

Thanks! This is now done via commit cb8539f.

tacaswell · 2015-02-18T00:27:18Z

Other than the testing issues Eric pointed out, this looks good.

Could you also check the docstrings to make sure there are not any references left to this restriction?

tacaswell · 2015-02-18T03:27:23Z

lib/matplotlib/axes/_axes.py

@@ -5607,7 +5606,7 @@ def hist(self, x, bins=10, range=None, normed=False, weights=None,
        # basic input validation
        flat = np.ravel(x)
        if len(flat) == 0:
-            raise ValueError("x must have at least one data point")
+            return [], [], cbook.silent_list('No Patches')


Do we want do draw no patches or patches of zero size? On consideration I think patches of zero size might be a better option here so client code which expects a non-zero length list does not have to check that it is true. For example if the user passes in bin edges, they should get back a list of zeros telling them that there are no counts in any of the bins.

@efiring @mdboom @WeatherGod @pelson Thoughts, I think that this is an important API choice that needs to be made.

efiring · 2015-02-18T19:31:29Z

I see that np.histogram returns an array of zeros and an array of boundaries when given an empty input array. I think we should follow that logic, which implies that we should be drawing zero-height (or width) patches.

WeatherGod · 2015-02-18T19:35:24Z

Yeah, I think that was one of my contributions to numpy... so, I certainly
would agree with that assessment.

Have we fixed issues with autoscaling the axis when there are empty patches?

Ben Root

On Wed, Feb 18, 2015 at 2:31 PM, Eric Firing notifications@github.com
wrote:

I see that np.histogram returns an array of zeros and an array of
boundaries when given an empty input array. I think we should follow that
logic, which implies that we should be drawing zero-height (or width)
patches.

—
Reply to this email directly or view it on GitHub
#4119 (comment)
.

umairidris · 2015-02-18T23:53:32Z

I have attempted to address the feedback (via 161eb17):

hist now is consistent with the result of np.histogram (in fact uses np.histogram's return value). Thus, if a user passes in bin edges they now get an array of zeroes back accordingly.
There is now (a single) zero-height and zero-width patch being drawn. This should also address the concern of client code checking/requiring there there to be at least 1 patch.

tacaswell · 2015-02-18T23:59:32Z

I think it should return the correct number of patches, drawn at the correct places. The application I am thinking of is if you doing something like

vals, edges, patches = ax.hist([], bins=15, bottoms=range(N))
for v, p in zip(vals, patches):
    f(v, p)

Also, putting the patch at 0, 0 is probably not right thing to do because it can mess up auto-limiting.

I strongly suspect that the right thing to do is to just let the results from the np.hist([], ...) propagate through the rest of the code as-is.

umairidris · 2015-02-21T04:50:07Z

I've let the results of np.hist propagate through the code (after some minor adjustments to part of the code that needed non empty input) and set the patch to be zero size (via width of bar patch and linewidth of fill patch). Please have a look!

tacaswell · 2015-03-13T01:27:52Z

lib/matplotlib/axes/_axes.py

@@ -5761,6 +5762,8 @@ def hist(self, x, bins=10, range=None, normed=False, weights=None,
            for m, c in zip(n, color):
                if bottom is None:
                    bottom = np.zeros(len(m), np.float)
+                if input_empty:
+                    width = 0


Do we need this?

As per @efiring 's comment: "I think we should follow that logic, which implies that we should be drawing zero-height (or width) patches."

Without setting width to 0, the patches for bar have size 0.1 and thus an empty graph (as desired) is not drawn.

Before (width 0.1, non empty graph):

After (width 0, empty graph):

However, currently the code still calculates the original width of 0.1 and then overwrites it, I will send a commit which moves it into a more appropriate place (where width, dw and boffset are originally calculated) in a future commit if this is the desired result.

I think the parenthetical widths comment was for drawing horizontal bars and that the example with the line is correct.

Done via 40cf945

tacaswell · 2015-03-15T23:02:40Z

We probably do not need the pdf and svg tests in this case. Can you please add extensions=['png'] to the kwarg of the decorator and re-base this branch to remove the pdf/svg from history?

umairidris · 2015-03-16T07:20:30Z

We probably do not need the pdf and svg tests in this case. Can you please add extensions=['png'] to the kwarg of the decorator and re-base this branch to remove the pdf/svg from history?

Done.

…ng for hist. Fix ambiguous truth value in hexbin.

…step on empty input.

… hist and hexbin to use baseline images.

umairidris · 2015-03-17T20:39:18Z

There were far too many useless commits in this PR (PEP8 fixes, comment/docstring changes, etc) given the relative size of the change (forgive my inexperience, this was my first PR). I've squashed some of the unnecessary commits together via rebase for a cleaner merge with master. Thanks!

tacaswell · 2015-03-18T04:07:46Z

@jenshnielsen This is having strange issue with the docs, can you take a look?

jenshnielsen · 2015-03-18T06:30:08Z

Looks like pip install of the doc dependencies just failed (the connection was cut in the middle of the IPython download). That really should make the build fail earlier. I have restarted the docs build

tacaswell · 2015-03-18T16:31:56Z

lib/matplotlib/axes/_axes.py


        # Massage 'x' for processing.
        # NOTE: Be sure any changes here is also done below to 'weights'
-        if isinstance(x, np.ndarray) or not iterable(x[0]):
+        if input_empty:
+            x = np.array([[]])


Why a 2D array instead of 1D?

never mind, I see why.

…nd not overwrite weights on empty input.

ENH : Allow empty input to hist and hexbin closes #3886

tacaswell · 2015-03-30T03:51:49Z

@umairidris Thank you!

Umair Idris added 3 commits February 17, 2015 18:10

Fix hexbin raising ValueError on empty dataset

e864d94

Fix hist raising ValueError on empty dataset

287a815

Add explanation for test_hist_empty

6deba68

tacaswell added the status: needs review label Feb 17, 2015

umairidris mentioned this pull request Feb 17, 2015

behavior when plotting no data #3886

Closed

umairidris changed the title ~~Fix ValueError being raised when plotting hist and hexbin on empty dataset (Issue #3886)~~ Fix ValueError being raised when plotting hist and hexbin on empty dataset Feb 17, 2015

efiring reviewed Feb 17, 2015
View reviewed changes

tacaswell added this to the next point release milestone Feb 18, 2015

tacaswell added status: needs revision and removed status: needs review labels Feb 18, 2015

tacaswell reviewed Feb 18, 2015
View reviewed changes

tacaswell added status: needs review and removed status: needs revision labels Feb 21, 2015

tacaswell reviewed Mar 13, 2015
View reviewed changes

Umair Idris added 3 commits March 17, 2015 16:04

Fix test_hist_empty and test_hexbin_empty test cases. Improve docstri…

74a4266

…ng for hist. Fix ambiguous truth value in hexbin.

Fix hist to return correct patches on empty input. Add test for hist …

a567658

…step on empty input.

Fix width and weights for hist on empty input. Update tests for empty…

2bf9777

… hist and hexbin to use baseline images.

tacaswell reviewed Mar 18, 2015
View reviewed changes

Fix hexbin to use len instead of any. Fix hist to pass in bin_range a…

22c6c7f

…nd not overwrite weights on empty input.

umairidris changed the title ~~Fix ValueError being raised when plotting hist and hexbin on empty dataset~~ Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix ##3886) Mar 23, 2015

umairidris changed the title ~~Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix ##3886)~~ Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix #3886) Mar 23, 2015

tacaswell added a commit that referenced this pull request Mar 30, 2015

Merge pull request #4119 from umairidris/hist_hexbin_empty

b7df247

ENH : Allow empty input to hist and hexbin closes #3886

tacaswell merged commit b7df247 into matplotlib:master Mar 30, 2015

tacaswell removed the status: needs review label Mar 30, 2015

umairidris deleted the hist_hexbin_empty branch April 12, 2015 03:35

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix #3886) #4119

Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix #3886) #4119

umairidris commented Feb 17, 2015

efiring Feb 17, 2015

umairidris Feb 18, 2015

tacaswell commented Feb 18, 2015

tacaswell Feb 18, 2015

tacaswell Feb 18, 2015

efiring commented Feb 18, 2015

WeatherGod commented Feb 18, 2015

umairidris commented Feb 18, 2015

tacaswell commented Feb 18, 2015

umairidris commented Feb 21, 2015

tacaswell Mar 13, 2015

umairidris Mar 13, 2015

tacaswell Mar 13, 2015 via email

umairidris Mar 14, 2015

tacaswell commented Mar 15, 2015

umairidris commented Mar 16, 2015

umairidris commented Mar 17, 2015

tacaswell commented Mar 18, 2015

jenshnielsen commented Mar 18, 2015

tacaswell Mar 18, 2015

tacaswell Mar 18, 2015

tacaswell commented Mar 30, 2015

Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix #3886) #4119

Fix ValueError being raised when plotting hist and hexbin on empty dataset (Fix #3886) #4119

Conversation

umairidris commented Feb 17, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tacaswell commented Feb 18, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

efiring commented Feb 18, 2015

WeatherGod commented Feb 18, 2015

umairidris commented Feb 18, 2015

tacaswell commented Feb 18, 2015

umairidris commented Feb 21, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tacaswell Mar 13, 2015 via email

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tacaswell commented Mar 15, 2015

umairidris commented Mar 16, 2015

umairidris commented Mar 17, 2015

tacaswell commented Mar 18, 2015

jenshnielsen commented Mar 18, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tacaswell commented Mar 30, 2015