I wonder if that belongs to the scope of this PR, but I think that the description used for the bins parameter in ax.hist may be outdated. If I am correct, the related keyword argument will be passed to numpy.histogram, which means that any string among {"auto", "fd", "doane", "scott", "rice", "sturges", "sqrt"} should be valid provided that the Numpy version is recent enough, right? Currently, it looks like only "auto" is mentioned.

For example, the following (inspired from the numpy docs) runs fine on my laptop with Matplotlib 2.1.0 and numpy 1.13 (both from conda)

import numpy as np
import matplotlib.pyplot as plt

prng = np.random.RandomState(10)  # deterministic random data
a = np.hstack((prng.normal(size=1000),
               prng.normal(loc=5, scale=2, size=1000)))

non_auto_bins_str = ("fd", "doane", "scott", "rice", "sturges", "sqrt")

fig, axs = plt.subplots(nrows=2, ncols=3, sharex=True)
for ax, bins in zip(axs.flat, non_auto_bins_str):
    ax.hist(a, bins=bins)  # arguments are passed to np.histogram
    ax.set_title("'{}' bins".format(bins))

fig.tight_layout()
fig.show()

I am hesitant to document the auto-binning on histogram more than 'with a new enough version of numpy strings will work, see np.histogram' until our minimum version of numpy is high enough that we are sure it will work.

How about: 'with a new enough version of numpy, strings that specify a binning strategy are also accepted, see np.histogram'.

fwiw that's numpy 1.11.

Thanks for the hint on the binning strategy. I've now written.

With Numpy 1.11 or newer, you can alternatively provide a string describing a binning strategy, such as ‘auto’, ‘sturges’, ‘fd’, ‘doane’, ‘scott’, ‘rice’, ‘sturges’ or ‘sqrt’, see numpy.histogram.

I think that's clear and useful, and without the danger that it will not work unexpectedly.