How do I specify the file path for figures rendered using Python?

[marcdotson]

Description

Figures rendered using R can have their file path specified using:

knitr:
  opts_chunk:
    fig.path: "../figures/"

What is the equivalent for figures rendered using Python?

[marcdotson]

This was answered for R in #2767.

[mcanouil]

It would depends but most Python libraries rely on matplotlib so tweaking the default configuration might work: savefig.directory https://matplotlib.org/stable/api/matplotlib_configuration_api.html#matplotlib.rcParams

[marcdotson]

It looks like savefig.directory works explicitly with saving figures. I'm looking for any figures made while rendering to end up in a specific directory, like with the fig.path option in R, rather than in a filename_files directory in the same location as the .qmd I'm rendering.

FWIW, here's a simplification of the Python code, including a figure, that I'm rendering in the Quarto document:

import numpy as np
import polars as pl
import matplotlib as pt
from matplotlib import pyplot as plt

# Set the default save directory
pt.rcParams['savefig.directory'] = '../figures'

np.random.seed(42)

# Set the parameter values.
beta0 = 3
beta1 = 7
n = 100

sim_data = (
    # Simulate predictors using appropriate np.random distributions.
    pl.DataFrame({
        'x': np.random.uniform(0, 7, size = n)
    })
    # Use predictors and parameter values to simulate the outome.
    .with_columns([
        (beta0 + beta1 * pl.col('x') + np.random.normal(0, 3, size = n)).alias('y')
    ])
)

sim_data

plt.scatter(sim_data['x'], sim_data['y'])

[mcanouil]

Jupyter/IPYthon does not offer this as far as I know. (images are base64 encoded inside the notebooks)
For knitr, the option comes directly from it, not from Quarto.

For now, Quarto sets some default using options such as fig-height, etc.

• quarto-cli/src/resources/jupyter/lang/python/setup.py

  Lines 20 to 29 in b822c3f

  	try:
  	import matplotlib.pyplot as plt
  	plt.rcParams['figure.figsize'] = (fig_width, fig_height)
  	plt.rcParams['figure.dpi'] = fig_dpi
  	plt.rcParams['savefig.dpi'] = "figure"
  	from IPython.display import set_matplotlib_formats
  	set_matplotlib_formats(fig_format)
  	except Exception:
  	pass

Unless explicitly saving the image, I don't see a way to change the default for Jupypter (Python or Julia).

This is the logic to come from base64 encoded images to file:

• quarto-cli/src/core/jupyter/jupyter.ts

  Lines 1943 to 1966 in b822c3f

  	if (
  	// base64 decode if it's not svg
  	mimeType !== kImageSvg ||
  	// or if it is encoded svg; this could happen when used in embed context,
  	// as Pandoc will generate ipynb with base64 encoded svg data
  	// https://github.com/quarto-dev/quarto-cli/issues/9793
  	!/<svg/.test(imageText)
  	) {
  	const imageData = base64decode(imageText);
  	// if we are in retina mode, then derive width and height from the image
  	if (
  	mimeType === kImagePng && options.figFormat === "retina" && options.figDpi
  	) {
  	const png = new PngImage(imageData);
  	if (
  	png.dpiX === (options.figDpi * 2) && png.dpiY === (options.figDpi * 2)
  	) {
  	width = Math.round(png.width / 2);
  	height = Math.round(png.height / 2);
  	}
  	}
  	Deno.writeFileSync(outputFile, imageData);
  	} else {

Note that if the Quarto team decides to do something here, it will have to work for all engines thus be a language agnostic option.

[marcdotson]

A language-agnostic option would be incredible. Is this worth submitting as an issue?

[mcanouil]

Actually, "A language-agnostic option" is not really possible in the sense of maintainability across all possible Jupyter kernels.
Quarto already made some efforts to offer control on how figures are defined/generated for some general libraries in Python, but a Quarto option would also need to work with the Julia engine and the Julia kernel.
Note that kernels can do whatever they want regardless of options Quarto might defines.