boxplot: positions used to take Int64Index · Issue #10022 · matplotlib/matplotlib · GitHub

Bug report

I upgraded matplotlib recently and my boxplot stopped working. I realized it was the keyword positions which was taking in a Int64Index array created by pandas. Using Int64Index.values fixes this, but plot takes a Int64Index array so can't see why it isn't desirable for boxplot or why it was removed/no longer works.

Code for reproduction

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
data = np.random.rand(5,2)
years = pd.date_range('1/1/2000',
                      periods=2, freq=pd.DateOffset(years=1)).year                      
# Does not work
plt.figure()
plt.boxplot(data, positions=years)
# Works
plt.figure()
plt.boxplot(data, positions=years.values)
# Works
plt.figure()
plt.plot(years, data[0,:])

Actual outcome

Traceback (most recent call last):

  File "<ipython-input-38-2540b7c8b9eb>", line 6, in <module>
    plt.boxplot(data, positions=years)

  File "/Users/Ray/anaconda/lib/python3.6/site-packages/matplotlib/pyplot.py", line 2729, in boxplot
    zorder=zorder, data=data)

  File "/Users/Ray/anaconda/lib/python3.6/site-packages/matplotlib/__init__.py", line 1717, in inner
    return func(ax, *args, **kwargs)

  File "/Users/Ray/anaconda/lib/python3.6/site-packages/matplotlib/axes/_axes.py", line 3378, in boxplot
    manage_xticks=manage_xticks, zorder=zorder)

  File "/Users/Ray/anaconda/lib/python3.6/site-packages/matplotlib/axes/_axes.py", line 3698, in bxp
    cap_x = np.array([cap_left, cap_right])

ValueError: setting an array element with a sequence.

Expected outcome

Expect to produce boxplot and label the xaxis '2000, 2001'
Unfortunately I cannot remember which version I previously used...

Matplotlib version

Operating system: macOS
Matplotlib version: 2.1.1
Matplotlib backend (print(matplotlib.get_backend())): module://ipykernel.pylab.backend_inline
Python version: Python 3.6.3 :: Anaconda custom (x86_64)
Pandas version: 0.21.1

Installation

conda upgrade matplotlib

p.s. don't have much dev experience but happy to have a crack at this if it is desired.

I think this is #9771. Pandas no longer registers their converter...

Thanks. I wasn't aware of the change in pandas.
It may be slightly different as it works for plot but not boxplot?
I tried adding the

from pandas.tseries import converter
converter.register()

to the top of my code but it didn't work. Am I missing something?

Well, not sure what the problem is then. Sorry for the misdirect...

But just to be sure, did you upgrade pandas at the same time? If so, I'd check if the change was on their end...

Yeah, I think I upgraded a few packages at the same time. I'm using pandas: 0.21.1 (added this to my summary). I'll update if I find out more info

The issue is likely that Index objects used to act more like arrays than they do now or we are now using an array-like interface that Index does not support, or numpy is doing better and preserving types(would have to trace through the code to be sure of who broke who which is probably not worth doing).

It looks like the exact issue is that

matplotlib/lib/matplotlib/axes/_axes.py

Lines 3680 to 3686 in 7a6bab7

    
           # width 
        
           if widths is None: 
        
               widths = [np.clip(0.15 * np.ptp(positions), 0.15, 0.5)] * N 
        
           elif np.isscalar(widths): 
        
               widths = [widths] * N 
        
           elif len(widths) != N: 
        
               raise ValueError(datashape_message.format("widths"))

is how we compute the box widths if position is passed in and we expect a list of floats, but for the Index types it is is returning a list of Index types (because np.ptp returns an Index type with one value in it). This makes sense for Index types in the context of pandas, but seems to be causing problems at the line the stack trace is on when we try to make an array out of the result of +/- operations on the Index object (which is still and Index object).

Just passing this to float does not work (float blows up), adding in a call to np.asanyarray might do the trick though...

boxplot: positions used to take Int64Index · Issue #10022 · matplotlib/matplotlib · GitHub

Repository navigation

boxplot: positions used to take Int64Index #10022

Bug report

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	# width
	if widths is None:
	widths = [np.clip(0.15 * np.ptp(positions), 0.15, 0.5)] * N
	elif np.isscalar(widths):
	widths = [widths] * N
	elif len(widths) != N:
	raise ValueError(datashape_message.format("widths"))

boxplot: positions used to take Int64Index · Issue #10022 · matplotlib/matplotlib · GitHub

Repository navigation

boxplot: positions used to take Int64Index #10022

Description

Bug report

Activity

jklymak commented on Dec 16, 2017

raybellwaves commented on Dec 16, 2017

jklymak commented on Dec 16, 2017

jklymak commented on Dec 16, 2017

raybellwaves commented on Dec 16, 2017

tacaswell commented on Dec 17, 2017

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions