objects.Hist: how to get a log scale for counts?

[tburnett]

I'm trying to duplicate the sns.histplot log_scale option to scale the counts. Or log=True in plt.hist.
The histogram disappears after a Plot.scale(y='log') .
Before and after:

[thuiop]

Yes, this is a known issue ; has to do with how the coordinates system works (notice there is no "0" in the log scale). See #2454 for more details. In the meantime, you should avoid using Plot.scale for this specific case, and use ax.set_yscale("log") manually.

[thuiop]

Depends on what you want to do exactly with the legend.

[tburnett]

Legend: In matplotlib, one has a "label=..." for each line or hist, and a fig.legend() collects them all.

import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(4,2))
ax.hist(r, np.linspace(-3,3,25), density=True, label='random.normal')
ax.plot(x,y, label='norm(0,1)')
fig.legend()

[thuiop]

Ah, I see. This is probably pretty complex to do ; I would probably just add a separate legend in that case.

[tburnett]

Not sure what you mean by "separate". Again, if I had a private way to produce a MPL-like legend, I'd continue exploring how to apply this system to my every-day efforts. I've already understood how to make a Plot subclass, which I use to get a log scale. (Minor, but It would be nice it Plot.clone could return an instance of its actual class.)
Related question: is there an example somewhere showing how to duplicate the MPL errorbar display?

[thuiop]

The legend seaborn creates is a matplotlib (Figure) legend. The only issue is that it cannot be modified afterwards without doing some weird stuff. However, matplotlib does support having several legends, so you can make a second "box" with the line you drew afterwards for instance.

[mwaskom]

Dupe of #3032?

[tburnett]

It doesn't look like it. Adding a "a. scale(y='symlog')" does not work. This is a very common use case in physics and astrophysics, overlaying a distribution with a function or functions. Publication-quality plots demand legends.

…

--Toby Burnett

________________________________ From: Michael Waskom ***@***.***> Sent: Monday, June 26, 2023 09:35 To: mwaskom/seaborn ***@***.***> Cc: Toby Burnett ***@***.***>; Author ***@***.***> Subject: Re: [mwaskom/seaborn] objects.Hist: how to get a log scale for counts? (Discussion #3401) Dupe of #3032<https://urldefense.com/v3/__https://github.com/mwaskom/seaborn/issues/3032__;!!K-Hz7m0Vt54!k59TO8El5jjoDGk8v7B5bkJxSyOVJN-sKTxF4ZI5y5ghiQbdn7WOFwzG3Ozl7TvIVrrdq-AMlKxFs-MciA1QxpY$>? — Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/mwaskom/seaborn/discussions/3401*discussioncomment-6284267__;Iw!!K-Hz7m0Vt54!k59TO8El5jjoDGk8v7B5bkJxSyOVJN-sKTxF4ZI5y5ghiQbdn7WOFwzG3Ozl7TvIVrrdq-AMlKxFs-McsRMNekw$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AAKXRPOMGOSC5JFT5V4TAE3XNG25HANCNFSM6AAAAAAZRVSE7Q__;!!K-Hz7m0Vt54!k59TO8El5jjoDGk8v7B5bkJxSyOVJN-sKTxF4ZI5y5ghiQbdn7WOFwzG3Ozl7TvIVrrdq-AMlKxFs-MckWcZE-A$>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[mwaskom]

Your post does not include a reproducible example so I really have no clue what your actual issue is. Telling me that it’s important in physics is pointless if you can’t actually articulate the problem.

[tburnett]

OK,

import seaborn.objects as so
import numpy as np
from scipy.stats import norm

r = np.random.normal(size=100000)
x = np.linspace(-4,4,100)
y = norm.pdf(x)

p =(so.Plot()
     .layout(size=(4,2)).limit(x=(-4,4))
     .add(so.Bars(), so.Hist(stat='density', bins=25), x=r)
     .add(so.Line(color='red'), x=x, y=y)
)

p.show(), p.scale(y='log').show()