PR Summary

I had a lot of bugs on the labelling on logit scale. For example, the following code:

import numpy as np 
import scipy as sp 
import scipy.stats 
import matplotlib.pyplot as plt 
 
s = np.linspace(-5,5,10000) 
s2 = np.linspace(-10,10,10000) 
lims = [(1e-10,1-1e-10),(1e-5,1-1e-5),(11e-2,15e-2)] 
for k,lim in enumerate(lims): 
    fig,ax = plt.subplots() 
    ax.set_yscale('logit') 
    ax.plot(s,sp.stats.norm.cdf(s)) 
    for i in range(9): 
        ax.plot(s2,sp.stats.t(i+1).cdf(s2)) 
    ax.grid() 
    ax.set_ylim(*lim) 
    fig.savefig(f'/tmp/g{k}.png')

Actual behavior

I obtain for the very large zoom:

For normal zoom:

For zoomed axis:

Behavior on the proposed branch

All this graphs are obtained with the same code as above.

I change the locator to allow different behavior, on large zoom, some decade are not proposed on majors, but on minors. No minor ticks used for subdivision of decades. Label are present only on major ticks. I obtain:

On normal zoom, all decades are shown on major, minor are indicating subdecade. Most of the time only major ticks are labelled, but when only a few of major are present, a spaced subset of minors can be labelled. I obtain :

On a very zoomed graph, the logiscale is kind-of linear, so the MaxNLocator is used (in particular, my LogitLocator inherit from MaxNLocator to allows fallback to MaxNLocator in this case, I obtain:

Also, two little things:

• for ticks labelling, the precision of format is adaptative, and depends of the diff between the labelled tick and neighbors,
• formater have use_oveline method, which allow use the survival notation (\overline{x} mean 1-x). Not enabled by default.

Should solve #13698.

PR Checklist

I putted my comments on each item.

• Has Pytest style unit tests
  • I tried to test the most critical behaviors of my code.
• Code is Flake 8 compliant
• New features are documented, with examples if plot related
  • Except for some new parameters, I did not add features.
• Documentation is sphinx and numpydoc compliant
  • I tried, but I did not test, and I don't know how to test that.
• Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
  • I don't know if this change is major. I think it is not.
• Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way
  • No backward incompatible API changes.