Interesting feature, and I definitely support the idea because the user has no other way to scale the histogram before display. More work will be needed on this PR, however, to get it up to snuff.

• Need to update pyplot.py using the boilerplate.py code (this should be done last)
• Need to add a note to the docstring for which version this kwarg was added (use .. versionadded::)
• Need to add an entry to doc/users/whats_new.rst for this feature (or, if more appropriate, doc/api/api_changes.rst)

Also desirable (let us know if you need help with either of these):

• Add a test for this.
• Add an example usage for the gallery.

Sure, I completely understand that more work is necessary. I largely opened to the request to see if people thought this would be a reasonable idea. I'll try to put some time in tonight to bring this into a mergeable form.

Is there any way to runboilerplate.py without installing the tree? It's currently giving meAttributeError: type object 'Axes' has no attribute 'hist2d', which suggests it's trying to run against my installed matplotlib-1.1.1.

No, it.assumes that whatever install of mpl is the one you are using to
rebuild pyplot.py.

If log is done this way, will the colorbar label come out right?. I mean, will the label show the value or log of value?

Ooh, good point, I hadn't thought of that. It would show, by default the
log of the values. Presumedly, one would set a log norm object in case the
user does not provide one for the scalar mappable. This would override the
default of Normalize for the ScalarMappable.

I believe I touched on most of the changes mentioned above in this updated series. Things have changed quite a bit,

• Instead of taking the log myself, I leave this forpcolorfast by passing it aLogNorm as thenorm kwarg (meaningcolorbar does the right thing, showing ticks labelled as 10^n in the logarithmic case)
• Instead of takinglog(x+1) I mask out the bins with zero counts asNone (or NaN, as this is apparently represented as)
• I added a default rc value for thehist2d.log option
• I added a brief snippet towhats_new.rst
• I ranboilerplate.py

How does it look to others?

Here is a quick patch adding a note pointing out how to use a logarithmic color scale. It might also be nice to demonstrate how to do this inhist2d_demo.py, perhaps by adding a second subplot? Unfortunately, it seems thathist2d_demo.py only uses pylab which apparently doesn't exposeLogNorm in any way. Would it be frowned upon to importLogNorm directly frommatplotlib.colors?

Any opinion on the above? It might be nice to see a patch along these lines go in to 1.2.

Any thoughts concerning adding a logarithmic histogram tohist2d_demo.py?

That would be a great idea.

Note, that I think it would be best to have it as a separate example

Alright, that is fine. I'll submit a patch with a new example shortly but do you suppose 3f2cdaf could be merged in the meantime?

How does this look as an example?

I think this is ready for merging. Any other requests?

Good catch. Fixed.

This pull requestfails (mergedd33877c intocf7618c).

Not entirely sure what@travisbot is upset about but I don't see anything that could be from my (quite trivial) work.

Don't fret. We are still working on configuring our automated testing
setup over at Travis. It has been complaining about everyone's commits the
past couple of days.

Good work@bgamari, I will merge.