Error autoscaling histogram with histtype='step'

[joshualande]

I am running into problems where histograms are not autoscaling correctly.

Below is a very simple example which reproduces this problem:

import pylab as P
P.hist([3000,3010, 3012], histtype='step')
P.savefig('test.pdf')

When I run this example using matplotlib v1.1.0 or using the absolute latest version of matplotlib, I obtain an axes which varies from 3000<x<3012 and 0<y<3000 with no visible histogram.

When I add (before saving the plot) the line:

P.gca().set_ylim(ymax=2.1)

I get a reasonable axes range and can see the histogram.

I also get a reasonable axes range when I remove the command histtype='step' or when I use matplotlib v1.0.0.

Thanks for your help,

Joshua

[taldcroft]

I see a (probably) related problem with the normed option and histtype='step' (or stepfilled).

% ipython --pylab

In [1]: hist(normal(100, 15, 10000), normed=True, histtype='step')
Out[1]: 
(array([  1.68129565e-05,   1.68129565e-04,   1.41228835e-03,
         7.06144175e-03,   1.79730506e-02,   2.50933376e-02,
         2.09405374e-02,   9.12943541e-03,   1.96711592e-03,
         3.02633218e-04]),
 array([  33.23287027,   45.12845802,   57.02404577,   68.91963352,
         80.81522128,   92.71080903,  104.60639678,  116.50198453,
        128.39757228,  140.29316004,  152.18874779]),
 <a list of 1 Patch objects>)

This produces a plot with y-axis from 0 to 40. Without histtype='step' I get the expected plot with y limits from 0 to 0.030.

I've seen this with matplotlib 1.1.0 on CentOS-5 x86_64 and matplotlib 1.0.1 on Snow leopard. I think that 1.0.0 was OK (or maybe 0.99.1), since I have old code with histtype that used to work.

[ycopin]

I guess I found the bug: in matplotlib/axes.py:7729 (for matplotlib 1.1.0), it says:

minimum = min(bins)

which is just wrong. Replace this line with something like:

minimum = min( min(m) for m in n )

fix the bug.

BTW, few lines below, the code:

                for m in n:
                    ymin = np.amin(m[m!=0]) # filter out the 0 height bins

does not make much sense either, but I have no idea what it is supposed to do anyway...

[ycopin]

Could this be corrected for matplotlib 1.2?

[tacaswell]

This looks like at has come back (the code from first report is still generates a graph with strange limits).

@neggert Can you take a look at this?

[neggert]

Sure. There's another bug a colleague pointed out to me that I've been wanting to look at, so I can work on these both tomorrow.

[neggert]

@tacaswell I can't seem to reproduce this on either v1.3.x or master. I get a plot with a y-axis that goes from 0 to 1 (which isn't great either because you can't see the line, but it's not the same issue).

[tacaswell]

Yeah, I didn't read carefully enough (parsed 'bad limits', not what sort of bad limits).

This looks like it is actually an issue with the generic auto-limit code. Sorry for the noise.

[mdboom]

Seems like the original issue has been fixed along the way. Closing.