[ENH]: Specifying figure size based on width and ratio/aspect

Problem

While playing around with style files, I realize that for a paper it often makes sense to specify the width (set by the publisher) and ratio (determined by what we plot). Although one can of course read up and remember the figure width etc, it is quite convenient to just use a style file to determine the width and then set the ratio when creating the figure.

Proposed solution

There is https://matplotlib.org/stable/api/_as_gen/matplotlib.figure.Figure.set_figheight.html so one way is of course to create the figure using the style file and then set the height suitably.

However, some options are:

set_figratio, basically:

set_figheight(self.get_figwidth()*ratio)

Add a figratio keyword to Figure (and subplots and figure and ...) that takes the width, either from figsize or add a new rcParam for figwidth. Not sure which is less confusing...

(I think one can argue that the width should be kept when modifying the ratio based on the publisher-defined argument.)

Edit: searching a bit more I found https://matplotlib.org/stable/api/figure_api.html#matplotlib.figure.figaspect which does the exact opposite to what I want. Maybe an option is to come up with a good name for a similar function that determines the size with fixed width? Although the original idea was to make it "simple".

Edit: searching a bit more I found https://matplotlib.org/stable/api/figure_api.html#matplotlib.figure.figaspect which does the exact opposite to what I want. Maybe an option is to come up with a good name for a similar function that determines the size with fixed width? Although the original idea was to make it "simple".

Could we just add a parameter to figaspect so the user can choose whether the height or width is modified?

Yeah, that would be a good way to deal with it! (Although I also hope that there can be a "simpler" way added to sort it out directly.)

Ah, I missed that figaspect only returns the height and width and doesn't do anything to a figure.

I believe a new standalone function would be suitable. Repurposing figaspect does not work because we'd end up with weird naming. Suggestion:

def figsize(*, width=None, height=None, aspect=None):
    ...

where exactly two of the three parameters must be given. Usage looks like

plt.figure(figsize=figsize(width=6, aspect=2))

Pro: This keeps the interfaces clean and simple
Con: It's a bit lengthy and (like with figaspect) one has to import the function.

I believe we can bear the cons. I'm inclied to also discourage figaspect then in favor of figsize.

The alternative would be to pass the relevant information into the figure and do the width/height resolution there. I've not been able to come up with a good API for that (all my attempts were unreadable, clumsy or too clever: plt.figure(figsize=(width, None, aspect)) or plt.figure(figsize={"width": 2, "aspect": 2}) or plt.figure(figsize=(width, None)@aspect(2)))

@timhoffm

@timhoffm I hope we have vanishingly few users who can't do figure(figsize=(6, 6/2)) so I don't see the advantage of figure(figsize=figsize(6, aspect=2))

I think what @oscargus is after is the ability to se the default figsize in the style sheet, and then modify the height by an aspect_ratio kwarg so calls in the code would be figure(aspect=2).

I think the only objection to that proposal would be that figure(figsize=(6, 6), aspect=2) would be ambiguous. But if we were clear that the height would always be modified by the ratio, I think that could work. My gut reaction in such cases, however, is to just write my own wrapper around figure/subplots that does what I want.

@oscargus

I think what @oscargus is after is the ability to se the default figsize in the style sheet, and then modify the height by an aspect_ratio kwarg so calls in the code would be figure(aspect=2).

Correct! Something like:

import matplotlib.pyplot as plt
import matplotlib as mpl
import scienceplots

with mpl.style.context(["science", "ieee"]):
    fig, ax = plt.subplots(aspect=2)

with the rationale that it is often the width which is the parameter that should be set as "fixed" by the style file.

I see. Can we come up with a reasonable specification that can be passed via figsize? The problem I see with aspect is that it interacts with figsize figure(figsize=(6, 6), aspect=2) is logically inconsistent; on top of being ambiguous. Can we come up with an API that explicitly states "I want the figsize with this aspect via keeping width (or adjusting height)?

So I am not good at fine-tuning in aspect space - much more comfortable just using inches. So one possibility is to let figsize=4 mean the figure is 4" high, and uses the default width. The only slight confusion there is that the usual spec is (w, h) so having one element default to the second element is a bit strange. However, I do think the "fixed width, variable height" case is common enough to warrant doing something here.

For only setting figure width supporting figsize=(None, 4) would be enough and more clear than figsize=4. We can easily do that - put in a PR if you like.

But this is not the topic of the original request, which is asking for sizing to aspect.

@oscargus

Oh figsize=(None, 4) would be good, and could be symmetric for folks who want figsize=(6, None). @oscargus may have a strong preference for an aspect ratio here, but as I said I don't personally think in terms of aspect ratios

I can work on this issue!

@oscargus

@oscargus may have a strong preference for an aspect ratio here, but as I said I don't personally think in terms of aspect ratios

I think that would do. Ideally, I'd use aspect (as it is easier to guessimate that the aspect should be 2 than to try different heights when you are not really sure about the exact width), but (None, 4) is still better than not being able to modify the height without knowing (bothering to figure out) the exact width (as you really do not want to change it).

@arkapudota

Ok, let's allow figsize tuples with at most one of the two entries being None. @arkapudota do you want to work on this?

This does not restrict our ability to expand for a possible aspect definition later.

Yes I can work on this

I'll close this because it's difficult to define a reasonable API for aspects. Instead, let's focus on the simpler topic of partially setting width / height #31400.

Navigation Menu

Repository navigation

[ENH]: Specifying figure size based on width and ratio/aspect #28758

Problem

Proposed solution

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[ENH]: Specifying figure size based on width and ratio/aspect #28758

Description

Problem

Proposed solution

Activity

rcomer commented on Aug 27, 2024

oscargus commented on Aug 27, 2024

rcomer commented on Aug 27, 2024

timhoffm commented on Aug 27, 2024

jklymak commented on Aug 27, 2024

oscargus commented on Aug 28, 2024

timhoffm commented on Sep 5, 2024

jklymak commented on Sep 5, 2024

timhoffm commented on Sep 7, 2024

jklymak commented on Sep 7, 2024

arkapudota commented on Nov 7, 2024

oscargus commented on Nov 8, 2024

timhoffm commented on Nov 11, 2024

arkapudota commented on Dec 2, 2024

timhoffm commented on Mar 26, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions