Display size of matplotlib figures?

[tsm0913]

Hi all! Quick question. In older versions, it used to be that changing matplotlib figure dpi would change the render size on screen. Recently, it looks like render size and figure resolution were decoupled.

Is there any way to control render size now?

For context, my figures are rendered on the scale of a piece of paper (for scientific figure publication), and as a result show up tiny on screen.

Thanks so much! Marimo is incredible. Truly changed my life and my research.

[dmadisetti]

Thanks @tsm0913 ! Can we get a sense of how you are controlling the figure size? If you set the DPI and width explicitly, marimo should respect it

[tsm0913]

Hi @dmadisetti ! I am controlling it as follows (for example)

_fig = plt.figure(figsize = (1.5,1), dpi=1200)

to get a 1200 dpi 1.5 in x 1 in figure. When I show it on screen by doing (e.g.) mo.hstack([_fig]) in the bottom of the cell it looks really small. It used to be that by changing dpi the figure would look larger/smaller. but, I think in a recent update dpi was decoupled from display size. Is there any way to control display zoom, so that the figure can be rendered larger in the way dpi used to do this?

keeping the 1.5 in x 1 in is important because that let's me make the figures on the exact scale at which they'll appear on the page, making it easy to interface with Adobe Illustrator.

[daizutabi]

The change in behavior was introduced in PR #9144, which decoupled the render resolution (DPI) from the display size on the screen.

Background and Rationale

The primary goal of decoupling was to ensure that increasing the DPI (for sharpness/quality) doesn't cause the plot to become physically massive on the screen. This is also important for consistency with SVG rendering. In current marimo, we normalize the display size based on a reference of 100 DPI:

factor = dpi / 100
# metadata width = actual_pixel_width / factor

Proposed Direction

I understand the need to control the physical display size in the notebook, especially for figures intended for scientific publications.

A solution, similar to what was implemented in Altair (vega/altair#3163), is to introduce a scale_factor. This would allow users to scale the display size independently of both the figure's figsize and its dpi.

By introducing such a parameter, we can satisfy both requirements:

1. Keeping dpi as a control for "sharpness" without breaking the layout.
2. Providing a way to scale the actual display size on the screen.

While the logic is straightforward to implement in marimo's matplotlib backend, I'm not sure the best way to expose this setting to the user.

[georgeoshardo]

_fig = plt.figure(figsize = (1.5,1), dpi=1200)

@tsm0913 I was also doing something like this and noticed dpi stopped doing anything today.
In the meantime I'm using:

def display_at_dpi(path, fig):
    width_px = int(fig.get_figwidth() * fig.dpi)
    return mo.image(path, width=width_px)

which works well enough for now (I've only tried it with SVGs)!