no effect with jupyter.plot.resolution and jupyter.plot.zoom options

[alusiani]

Hi,

thanks for publishing this R Jupyter kernel! I have tried to use it, successfully, but I am getting too-small-sized ggplot inline plots in Jupterlab. I can surely increase the size of the plots using the options jupyter.plot.width and jupyter.plot.height, but that results in small points, line widths and characters, unless heavily customizing. With IRKernel, I can use option repr.plot.res to rescale the inline plots as I like. Looking at the source code of RKernel, it seems that I should use options jupyter.plot.resolution and jupyter.plot.zoom: I tried them but they apparently produce no effect at all. I attach a minimal notebook as example. Is there a way I could get larger inline plots than the default? I would like to ge the default ggplot plot text at the same size as the notebook cells text size.

library(ggplot2)

rc = ggplot(mpg, aes(displ, hwy, colour = class)) +
  geom_point()

options(jupyter.plot.width = 7, jupyter.plot.height = 7, jupyter.plot.resolution = 72, jupyter.plot.zoom = 1)
rc

options(jupyter.plot.width = 8, jupyter.plot.height = 4, jupyter.plot.resolution = 72, jupyter.plot.zoom = 1)
rc

options(jupyter.plot.width = 8, jupyter.plot.height = 4, jupyter.plot.resolution = 200, jupyter.plot.zoom = 1)
rc

options(jupyter.plot.width = 8, jupyter.plot.height = 4, jupyter.plot.resolution = 72, jupyter.plot.zoom = 2)
rc

[melff]

This should be fixed with the most recent commit 73150c2.

[alusiani]

This should be fixed with the most recent commit 73150c2.

Thanks for the reply and the update of RKernel.

I installed RKernel 1.1.1 and I see that jupyter.plot.zoom does something, while I don't see changes when modifying jupyter.plot.resolution. Apparently the set of options jupyter.plot.width = 8, jupyter.plot.height = 4, jupyter.plot.resolution = 288, jupyter.plot.zoom = 2 results in an effective ggplot2 device size of 8/2 inch x 4/2 inch. The user can then specify larger jupyter.plot.width, jupyter.plot.height. If the user wants a 8x4 inch^2 plot as saved on file, and wants to display the plot with a larger size on jupyter notebook, he should inflate jupyter.plot.width, jupyter.plot.height by a zoom factor and set jupyter.plot.zoom to the same factor. the above works but in my opinion is less convenient than the IRkernel repr.* options "width", "height" and "res", which would stay fixed to the desired ggplot2 width and height, while "res" can adjust the plot size in jupyter. I think that the same behaviour would be best for RKernel as well. My impression is that the "resolution" and "zoom" options are redundant, and that the "resolution" option should be enough, provided that it works like in IRkernel.

[melff]

The "jupyter.plot.resolution" affects only bitmap graphics. The default graphics format in RKernel is effectively svg, a vector format.

The graphics format is set by the "jupyter.plot.formats" option. Its default setting is c("png","svgp","pdf").
It appears that Jupyterlab prefers svg/svgp output, so if you really want to see bitmap graphics, use options(jupyter.plot.formats=c("png","pdf")). (I would not recommend dropping the pdf format if you want to export your notebook to PDF via LaTeX or otherwise.)

The options 'jupyter.plot.width' and 'jupyter.plot.height' are designed so that the dimensions are the same on any device. They will also affect the result of ggsave() if called without width and height arguments or with the same width and height as the jupyter.plot.width/height options.

[melff]

This is how the zoom option works right now:

But it might indeed more sensible to make the zoom option affect the overall size of the plot as it appears on screen.

[alusiani]

I move here an issue message that I wrongly posted to another issue thread. I observe that setting jupyter.plot.width/height/zoom to non integer values results in errors. I attach an example notebook showing the error I get. I attach a notebook that shows the error that I get: rkernel-tests.ipynb .