Removing inkscape package from the minimal-notebook

[Bidek56]

Describe your changes

Removing inkscape package from the minimal-notebook as suggested by @benz0li in this issue.

Checklist (especially for first-time contributors)

• [x ] I have performed a self-review of my code
• [ x] If it is a core feature, I have added thorough tests
• [x ] I will try not to use force-push to make the review process easier for reviewers
• [ x] I have updated the documentation for significant changes

[benz0li]

@Bidek56 For the r-notebook and the datascience-notebook you need to create file /opt/conda/lib/R/etc/Rprofile.site and add either

options(jupyter.plot_mimetypes = c('text/plain', 'image/svg+xml', 'application/pdf'))

or

options(jupyter.plot_mimetypes = c('text/plain', 'image/png', 'image/jpeg', 'image/svg+xml', 'application/pdf'))

depending on how many different formats the notebook file (.ipynb) should include.

[Bidek56]

Any reason to use /opt/conda/lib/R/etc/Rprofile.site instead of ~/.Rprofile? Thx

[Bidek56]

I do not know much about R but based on this article we can set the plot_mimetypes right in the code. Why would we want to set them globally for all users? Thx

[benz0li]

Why would we want to set them globally for all users? Thx

Setting c.InlineBackend.figure_formats = {"png", "jpeg", "svg", "pdf"} in jupyter_server_config.py sets it globally for the Python kernel (IPython).

Setting options(jupyter.plot_mimetypes = c('text/plain', 'image/png', 'image/jpeg', 'image/svg+xml', 'application/pdf')) in /opt/conda/lib/R/etc/Rprofile.site sets it globally for the R kernel (IRkernel).

[benz0li]

I am only using c.InlineBackend.figure_formats = {"svg", "pdf"} and options(jupyter.plot_mimetypes = c('text/plain', 'image/svg+xml', 'application/pdf')); no need for png and jpeg.

ℹ️ Every format specified is stored in the notebook file (.ipynb).

[Bidek56]

Why set these at all? Why not add a documentation line showing users how to add this line to their notebook: options(jupyter.plot_mimetypes = "image/svg+xml")

[benz0li]

Why set these at all? Why not add a documentation line showing users how to add this line to their notebook: options(jupyter.plot_mimetypes = "image/svg+xml")

Removing inkscape without setting these globally will result in errors by default. That should not be the case.

It must work with the default settings.

[Bidek56]

1. I don't like copy-pasting the profile file. I suggest we add it to minimal-notebook.

Are you saying that you are OK with adding /opt/conda/lib/R/etc/Rprofile.site, ~/.Rprofile or neither? Thx

[Bidek56]

Why set these at all? Why not add a documentation line showing users how to add this line to their notebook: options(jupyter.plot_mimetypes = "image/svg+xml")

Removing inkscape without setting these globally will result in errors by default. That should not be the case.

It must work with the default settings.

This code works without jupyter.plot_mimetypes:

# Create a plot with some normally distributed data
library("ggplot2")
set.seed(42)
n <- 1000
p <- ggplot(data.frame(x = rnorm(n), y = rnorm(n)), aes(x=x, y=y)) + 
        geom_point(alpha = 0.25, size = 1, colour = "blue") + geom_density2d(colour = "red")

p + ggtitle(sprintf("Mime type = '%s'", getOption("jupyter.plot_mimetypes")))

since by default it seems to be set to image/png and application/pdf according to the docs Can you please share a test case or code sample where it would fail without jupyter.plot_mimetypes? Thx

[mathbunnyru]

1. I don't like copy-pasting the profile file. I suggest we add it to minimal-notebook.

Are you saying that you are OK with adding /opt/conda/lib/R/etc/Rprofile.site, ~/.Rprofile or neither? Thx

Yes, I'm ok with both options. I prefer /opt/conda/ version, it is unlikely to mess with the user mounts.

Why set these at all? Why not add a documentation line showing users how to add this line to their notebook: options(jupyter.plot_mimetypes = "image/svg+xml")

Removing inkscape without setting these globally will result in errors by default. That should not be the case. It must work with the default settings.

This code works without jupyter.plot_mimetypes:

# Create a plot with some normally distributed data
library("ggplot2")
set.seed(42)
n <- 1000
p <- ggplot(data.frame(x = rnorm(n), y = rnorm(n)), aes(x=x, y=y)) + 
        geom_point(alpha = 0.25, size = 1, colour = "blue") + geom_density2d(colour = "red")

p + ggtitle(sprintf("Mime type = '%s'", getOption("jupyter.plot_mimetypes")))

since by default it seems to be set to image/png and application/pdf according to the docs Can you please share a test case or code sample where it would fail without jupyter.plot_mimetypes? Thx

Unfortunately, I can't help with the example here. @benz0li maybe you could provide one?

[benz0li]

The default for IPython is definitively c.InlineBackend.figure_formats = {"png"} only – plus text/plain, the non-removable default for IPython.

Tested with jupyter/scipy-notebook and

import numpy as np
import matplotlib.pyplot as plt

r = np.arange(0, 2, 0.01)
theta = 2 * np.pi * r
fig, ax = plt.subplots(
  subplot_kw = {'projection': 'polar'} 
)
ax.plot(theta, r)
ax.set_rticks([0.5, 1, 1.5, 2])
ax.grid(True)
plt.show()

ℹ️ Image size: 277 x 269 pixels, Image DPI: 72 pixels/inch (Retina would be Image size: 555 x 538 pixels, Image DPI: 144 pixels/inch)

Notebook file: No application/pdf in data; "text/plain": ["<Figure size 432x288 with 1 Axes>"]

[benz0li]

The default for IRkernel is jupyter.plot_mimetypes = c('text/plain', 'image/png') with jupyter.plot_scale = 2.

Tested with jupyter/r-notebook and

library(ggplot2)

ggplot(airquality, aes(Temp, Ozone)) + 
  geom_point() + 
  geom_smooth(method = "loess"
)

ℹ️ Image size: 840 x 840 pixels, Image DPI: 120 pixels/inch (Due to jupyter.plot_scale = 2 displayed as 480 x 480 pixels, 240 pixels/inch [aka retina] in JupyterLab)

getOption("jupyter.plot_mimetypes")

'text/plain' 'image/png'

getOption("jupyter.plot_scale")

2

Notebook file: No application/pdf in data; "text/plain": ["plot without title"]

[benz0li]

Hope this helps in understanding the current IPython and IRkernel defaults.

And as mentioned before: IJulia is a whole different story...

[mathbunnyru]

What's the status of this PR?

[Bidek56]

What's the status of this PR?

I am trying to create tests as per your request but unfortunately I do not know enough about it.

[Bidek56]

@benz0li @mathbunnyru I have added basic R tests, unfortunately I do not know enough about ggplot2 to add more advance tests. Feel free to add more sophisticated tests or share them so I can add them. Thanks

[mathbunnyru]

Could you please remove code duplication as well?

[Bidek56]

Which code duplication? In the tests?

[mathbunnyru]

The test file, Rprofile.site config file and some lines in Dockerfiles.

[Bidek56]

They are used in different stacks, how would you suggest we re-use them across multiple stacks?

[mathbunnyru]

They are used in different stacks, how would you suggest we re-use them across multiple stacks?

As we discussed above, for the config file, let's put it to minimal-notebook. With the test file, just create a test file somewhere with all the code and then import then create two small files in the notebook test folders importing this file content. I think this will work well. Or, put the code in one of two test files and import it in another one.

[Bidek56]

They are used in different stacks, how would you suggest we re-use them across multiple stacks?

As we discussed above, for the config file, let's put it to minimal-notebook. With the test file, just create a test file somewhere with all the code and then import then create two small files in the notebook test folders importing this file content. I think this will work well. Or, put the code in one of two test files and import it in another one.

The jupyter_server_config.py config file is in base-notebook, why would we move it to minimal? Rprofile.site belongs to datascience-notebook and r-notebook so it does not make senses to move to move to minimal, what I am missing here? Thx

[mathbunnyru]

Sorry for not being clear.

I propose the following:

1. jupyter_server_config.py stays in the same place - I do not propose to change it's location.
2. Rprofile.site should be in minimal-notebook. There are 2 good reasons for this - code duplication and also that this change might break the R users who for whatever reason decided to have their images based on minimal-notebook. This way we will try not to break these users.
3. minetypes_R.ipynb and test_R_notebook.py should not be duplicated as well - I described how we can avoid the duplication quite easily.

[benz0li]

With the current changes, IPython and IRkernel are not in sync regarding figure formats/plot mimetypes.

Either reduce c.InlineBackend.figure_formats to {"svg", "pdf"} or extend jupyter.plot_mimetypes to c('text/plain', 'image/png', 'image/jpeg', 'image/svg+xml', 'application/pdf'). 👉 Either https://github.com/jupyter/docker-stacks/pull/1765#issuecomment-1209768706 or https://github.com/jupyter/docker-stacks/pull/1765#issuecomment-1209773517.

ℹ️ Every format specified is stored in the notebook file (.ipynb).