%matplotlib inline does not exist

[samzhang111]

I realize this might be a metakernel issue, so feel free to redirect me if that is the case. I am trying to run a scientific Python stack with pandas, matplotlib, etc. with this notebook, but it seems like I can't plot inline.

Any directions on where to start looking? I'd be happy to help contribute a patch.

[dsblank]

I've thought about this problem a bit, and I do think the solution should go into metakernel so that all derived kernels could set matplotlib up. I would look to see exactly what the %matplotlib magic does, and then we can add hooks to do similar things in metakernel.

[samzhang111]

It does look like there is a %plot magic, which has an inline attribute. I tried to run it with Hy but it opened a Tk backend. Is there something we need to implement in the Hy-specific code after all to get this to work?

https://github.com/Calysto/metakernel/blob/master/metakernel/magics/plot_magic.py https://github.com/Calysto/metakernel/issues/10 (which you're a part of :)

[dsblank]

Yes, that is the API of how we could configure, say matplotlib in Hy. Currently, I think it is only used to configure plotting in Octave. All we need to do it have Hy "do the right thing" when the plot settings are made.

[samzhang111]

Do you think we'd be able to re-use some code, like the InlineBackend, from ipykernel? After spelunking in the ipython code for 20 mins I'm still not quite sure how matplotlib is made to not open a backend, and instead make the images accessible to the kernel.

I was looking at the octave kernel and it looks like there the images get written to disk, then read again.

It also seems like there's a bit of detail involved in making sure the plot settings get set correctly with the metakernel base class. Is this a high enough priority for you that you could give me more instructions, or if you want, just take it yourself :) ?

Edit: I've found that https://github.com/bollwyvl/hy_kernel is able to plot inline, so this isn't an urgent request

[dsblank]

Yes, I think reuse or a variation of ipykernel is the way to go. I got lost in there too. The other way to do this is to just figure out how to to it directly, and then make %plot do that as simply as possible. I'll have to look at @bollwyvl 's code... I didn't realize he had that working. That should be a good hint at what metakernel needs to do in general.

[dsblank]

BTW, here is an easy way to hook into IPython's magics:

get_ipython().magic(u'matplotlib inline')

More at: https://stackoverflow.com/questions/35145509/why-is-ipython-display-image-not-showing-in-output

[schuster-rainer]

I created a reader macro (this is for hy 0.13 and newer)

(import [IPython [get_ipython]]
        [IPython.terminal.interactiveshell [TerminalInteractiveShell]])

(cond
 [(not (TerminalInteractiveShell.initialized))
  (TerminalInteractiveShell.instance)])

(defsharp % [expr]
  `(.magic (get_ipython) '~expr))

with this you can execute your magics like this:

#% matplotlib

@dsblank maybe you can add it to the initialisation of calysto

[dsblank]

Sure! Is it possible for you to make a PR?

[schuster-rainer]

Yes ... and be aware of syntax issues. defsharp has been some intermediate name.

from defreader > defsharp > and now deftag in the current master.

[dsblank]

Because Calysto Hy is based on Metakernel, I think it has a mechanism to run start up code (in the kernel's language) when the kernel starts. See https://github.com/Calysto/metakernel/tree/master/metakernel for more information.

[CFiggers]

For whomever might still be concerned, I ran into this issue while trying to get Calysto Hy working today. I wasn't able to get a workaround going—I'm using the Jupyter extension for VS Code.