shifted(?) bin positions when plotting multiple histograms at the same time

[anntzer]

Consider

import numpy as np
from matplotlib import pyplot as plt

mu, sigma = 1500, 100
x = mu + sigma * np.random.randn(1000, 3)

_, (ax1, ax2) = plt.subplots(2)
ax1.hist(x[:, 0], 30, range=(0, 3000), width=100)
ax1.hist(x[:, 1], 30, range=(0, 3000), width=100)
ax1.hist(x[:, 2], 30, range=(0, 3000), width=100)
ax2.hist(x, 30, range=(0, 3000), width=100)
plt.show()

[tacaswell]

This is nominally a feature, if you make the bars narrower they will line up in groups of 3 per bin.

[anntzer]

Perhaps some sort of warning when width is explicitly set then? (It seems more appropriate to use rwidth in that case).

[tacaswell]

It looks like width is only doing something because it is falling through to the update call at the bottom of hist ( width is not explicitly documented and not really implicitly documented either as the kwargs point to patch properties and width is a Rectangle property).

If you don't pass width, it 'does the right thing':

import numpy as np
from matplotlib import pyplot as plt

mu, sigma = 1500, 100
x = mu + sigma * np.random.randn(1000, 3)

_, (ax1, ax2) = plt.subplots(2)
ax1.hist(x[:, 0], 30, range=(0, 3000))
ax1.hist(x[:, 1], 30, range=(0, 3000))
ax1.hist(x[:, 2], 30, range=(0, 3000))
ax2.hist(x, 30, range=(0, 3000))
plt.show()

Explicitly passing width in over-determines several things (like if you change bins or range, you can now pass in a width which is wrong).