setting yscale to log, after drawing a plot with values equal to zero, results in incorrect handling of zero values

mpl 1.5.3 and 2.0, py3.6

_xs = np.linspace(-50, 50, 1001); plt.semilogy(scipy.stats.norm.pdf(_xs))

results in

where zero values (due to floating point resolution limits, at the edges) get ignored, as expected, but

_xs = np.linspace(-50, 50, 1001); plt.plot(scipy.stats.norm.pdf(_xs))
plt.gcf().canvas.draw()
plt.gca().set(yscale="log")

results in the incorrect

I think this boils down to lines 3241-3242 in axes/_base.py:

if value.lower() == 'log' and 'nonposy' not in kwargs:
    kwargs['nonposy'] = 'mask'

and is fixed by changing 'mask' to 'clip'.

Working on it now.

Fixed by #8836

Navigation Menu