[MRG] set vmin, vmax after smoothing in plot_epochs_image

[mmagnuski]

Another small change to plot_epochs_image - I made it separate from the other PR, because it is not necessarily a bug.
Currently vmin and vmax are set before smoothing (using sigma) so if one uses sigma the colorscale is not fully used (smoothed data will have lower min and max values etc.).
You can see the difference here:
Before:

After:

⚠️
Now the vmin, vmax are not set based on the whole data, but based on the first smoothed channel (out of those in picks). I could move smoothing out of the figure-loop - so that all picks are smoothed before the loop and vmin, vmax are set based on the whole smoothed data.

[dengemann]

Looks pretty cool. Taking a look later.
On Wed, 29 Jun 2016 at 01:31, Mikolaj Magnuski notifications@github.com
wrote:

Another small change to plot_epochs_image - I made it separate from the
other PR, because it is not necessarily a bug.
Currently vmin and vmax are set before smoothing (using sigma) so if one
uses sigma the colorscale is not fully used (smoothed data will have
lower min and max values etc.).
You can see the difference here:
Before:
[image: before_02]
https://cloud.githubusercontent.com/assets/8452354/16435741/6e2c6ac4-3d98-11e6-9306-878f1e131455.PNG

After:
[image: after_02]
https://cloud.githubusercontent.com/assets/8452354/16435747/77e06b2e-3d98-11e6-8dff-c06687b1adee.PNG

⚠️
Now the vmin, vmax are not set based on the whole data, but based on the
first smoothed channel (out of those in picks). I could move smoothing out
of the figure-loop - so that all picks are smoothed before the loop and

vmin, vmax are set based on the whole smoothed data.

You can view, comment on, or merge this pull request online at:

#3360
Commit Summary

• set vmin, vmax after smoothing

File Changes

• M mne/viz/epochs.py
  https://github.com/mne-tools/mne-python/pull/3360/files#diff-0 (5)

Patch Links:

• https://github.com/mne-tools/mne-python/pull/3360.patch
• https://github.com/mne-tools/mne-python/pull/3360.diff

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#3360, or mute the thread
https://github.com/notifications/unsubscribe/AB0fik365srsXKntqixkCNikC2VxJf9Wks5qQa7MgaJpZM4JAovy
.

[codecov-io]

Current coverage is 86.60%

Merging #3360 into master will increase coverage by 0.04%

@@             master      #3360   diff @@
==========================================
  Files           335        335          
  Lines         57319      57628   +309   
  Methods           0          0          
  Messages          0          0          
  Branches       8713       8721     +8   
==========================================
+ Hits          49617      49908   +291   
+ Misses         5075       5030    -45   
- Partials       2627       2690    +63   

Powered by Codecov. Last updated by 11fc0ef...cbba5fe

[jona-sassenhagen]

Both this and #3359 are very appreciated.

[agramfort]

I don't like the fact that you only use the first image.

I would say that if vmin, vmax are not set I see 2 options:

• have a different vmin, vmax per channel
• precompute the smoothed data and set the vmin, vmax globally

no strong feeling

[jona-sassenhagen]

I prefer option 2 - the code is probably more complex, but IMO same range = least surprise.

[mmagnuski]

I also think I prefer option two. I'll add precomputing smoothed data.

[mmagnuski]

I could do the np.swapaxes in one line with epochs.get_data() and then set n_epochs = data.shape[1] if you prefer.

[mmagnuski]

@agramfort @jona-sassenhagen @dengemann
Travis error is not related (failure to download some dataset), ready for review.

[agramfort]

thx @mmagnuski

[mmagnuski]

thanks! 🚀

[jona-sassenhagen]

Thanks Mikolaj!