plot_dot

[canyon289]

Describe the bug
plotdot fig size doesn't behave the way I expect, in that when I set figsize in an axes that triple a previous one its not triple the size. There also are some minor bugs where the dots seem to be overlapping some

To Reproduce

samples = stats.beta(2,2).rvs(100)

width = 10
fig, ax = plt.subplots(figsize=(width, 10))
az.plot_dot(samples, ax=ax)
ax.set_title(f"Width: {width}")
ax.set_xlim(0,1)

Then try this, but see that figure is not three times the width

width = 30
fig, ax = plt.subplots(figsize=(width, 10))
az.plot_dot(samples, ax=ax)
ax.set_title(f"Width: {width}")
ax.set_xlim(0,1)

Expected behavior
Figsize from plt.subplots is respected

Additional context
Arviz '0.12.0'

[sethaxen]

The overlapping stacks appears to be an issue with wilkinson_algorithm:

>>> values = stats.beta(2,2).rvs(100)

>>> values.sort()

>>> binwidth = 0.07

>>> stack_locs, stack_count = az.plots.dotplot.wilkinson_algorithm(values, binwidth)

>>> np.diff(stack_locs) > binwidth  # should all be True
array([ True,  True,  True,  True, False,  True,  True,  True,  True,
        True,  True, False])

[sethaxen]

Ah, okay, overlapping stacks is actually not a bug. This is possible and intended behavior. e.g. suppose we have the data [1, 1.9, 2.1, 2.1, 3.2] and use a binwidth=1. Then the bins would be [[1, 1.9], [2.1, 2.1], [3.2]], the bin centers would be [1.45, 2.1, 3.2], and the bin edges would be [[0.95, 1.95], [1.6, 2.6], [2.7, 3.7]], so the first and second stacks will overlap. Not allowing overlap would require the first or second stack be further from the underlying data.

This is how ggplot2::geom_dotplot behaves:

> x <- c(1, 1.9, 2.1, 2.1, 3.2)
> df <- data.frame(x)
> ggplot(df, aes(x=x)) + geom_dotplot(binwidth=1)

[ahartikainen]

Good to know this. Then PR should only fix aspect

[canyon289]

Wow what an indepth analysis. Thanks @sethaxen