mpl1.5.3 and mpl2.0b4

semilogy([1, 2, 3], [1, np.nan, 2], "o")

triggers a warning:

/usr/lib/python3.5/site-packages/matplotlib/scale.py:93: RuntimeWarning: invalid value encountered in less_equal
  mask = a <= 0.0

due to the comparison that involves a nan. I think this warning should be masked out (likely with the application of a np.errstate at the right place), given that nans are (AFAIK) a fully supported mechanism for hiding a point in a plot, and e.g. do not raise any warning in linear scale.