search

mne-tools / mne-python

MNE: Magnetoencephalography (MEG) and Electroencephalography (EEG) in Python
https://mne.tools
BSD 3-Clause "New" or "Revised" License
2.62k stars 1.3k forks source link

Interactive plots with show=False #5108

Open bloyl opened 6 years ago

bloyl commented 6 years ago

Interactive plots with show=False or with non blocking matlabplot backends can lead to unexpected results.

For instance (here is gist showing this for sample data - https://gist.github.com/bloyl/7f6a7910bf2afce5f198b6233f39416b )

evoked.info['bads'] = []
evoked.plot(exclude=[], show=False)
evoked.info['bads'] += [new_bad_chans]
evoked,.plot(exclude=[],show=False)
plt.show()

This will make 2 figures. the Butterfly plots will be correct, the first will show no bad channels and the second will show the bads chans in red.

However if you select a time range from either plot you will see a topomap created using on the 'good' channels of evoked. This is unexpected for 2 reasons. 1) exclude=[] should use all the channels. Maybe indicating bad channels somehow. 2) in figure 1, the topomap doesn't match the butterfly plot because the data used to make it changed but only the interactive part (the topomap) saw that change.

This is understandably a corner case and not a high priority but I wanted it listed so I don't completely forget about it :)

larsoner commented 6 years ago

Yes I could see how this would happen. I'm not sure if there is a good solution, though. Should we add something to the plot_visualize_evoked or somewhere else about this behavior? This will also affect all plotting modes where you can modify an object (in weird/probably breaking ways), such as epochs.plot and raw.plot.

bloyl commented 6 years ago

I don't know enough about python to know if this is possible, but I would think the ideal case would be to monitor the object and then redraw derivative plots (butterfly, topos etc) if the object is modified in a meaningful way. An alternative would be store a copy of the object for plotting, but that would have huge overhead for epochs or raw.

On Mon, Apr 9, 2018 at 1:15 PM Eric Larson notifications@github.com wrote:

Yes I could see how this would happen. I'm not sure if there is a good solution, though. Should we add something to the plot_visualize_evoked or somewhere else about this behavior? This will also affect all plotting modes where you can modify an object (in weird/probably breaking ways), such as epochs.plot and raw.plot.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/mne-tools/mne-python/issues/5108#issuecomment-379826500, or mute the thread https://github.com/notifications/unsubscribe-auth/AFU7nb7BeQYdILo3eYY3AlQhZ4M3JdmKks5tm5cZgaJpZM4TM4K1 .

agramfort commented 6 years ago

at least

evoked.plot

could be fixed as it cannot change the evoked instance.

drammock commented 3 years ago

I don't know enough about python to know if this is possible, but I would think the ideal case would be to monitor the object and then redraw derivative plots (butterfly, topos etc) if the object is modified in a meaningful way. An alternative would be store a copy of the object for plotting, but that would have huge overhead for epochs or raw.

another option to consider here would be disabling interactivity of a figure when show=False. That way, the revealed plot would be accurate as of the time it was created, no (incorrect) derivative plots could be produced from it, and the lack of interactivity would sort of reflect its "static-ness".

@bloyl do you often want to initialize a hidden plot, run some other code, and then later call plt.show() and interact with the revealed plot? Or was this a case of "I did this by accident once, and I'm documenting it because it seems bad"?