MRG, ENH: Add example of projection to source space

[larsoner]

@SherazKhan asked if we had an example of projecting decoder coefficients to source space yet and I didn't think we did. This adds it to the decoding tutorial, and cleans up some refs to use the new footbib magic.

[dengemann]

The way it’s coded here you add back the mean (via inverting scaler) which gives you essentially the grand average ERF.
This could be fixed by resubtracting the scaler.mean_

[larsoner]

It seems weird to add in this manual step. Isn't it informative to start to know that it is basically making use of the grand average ERF to do the classification, even if it does some processing like mean subtraction and standardization, PCA, whatever else, along the way?

[dengemann]

But the model uses the de-meaned data. You essentially add back the intercept which will suffocate the model specific-patterns.

[larsoner]

Here I just make use of an existing evoked_time_gen that's already part of the narrative. So if we're going to make that change, it's actually orthogonal to this one (should be separate PR). Can you make a PR to improve this part of the tutorial and make evoked_time_gen more informative in this regard?

[dengemann]

Yeah it’s an old issue just seeing that brings it back to memory. I think we should make these considerations part of the tutorial and explore the different outputs.

…

On 30 Apr 2020, at 19:44, Eric Larson ***@***.***> wrote:  Here I just make use of an existing evoked_time_gen that's already part of the narrative. So if we're going to make that change, it's actually orthogonal to this one (should be separate PR). Can you make a PR to improve this part of the tutorial and make evoked_time_gen more informative in this regard? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[codecov]

Codecov Report

Merging #7705 into master will decrease coverage by 0.05%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #7705      +/-   ##
==========================================
- Coverage   90.35%   90.30%   -0.06%     
==========================================
  Files         455      455              
  Lines       84419    84561     +142     
  Branches    13369    13396      +27     
==========================================
+ Hits        76274    76360      +86     
- Misses       5273     5318      +45     
- Partials     2872     2883      +11     

[larsoner]

Updated doc:

https://19760-1301584-gh.circle-artifacts.com/0/dev/auto_tutorials/machine-learning/plot_sensors_decoding.html#projecting-sensor-space-patterns-to-source-space

It seems to look wacky because we are getting "403 (Forbidden)" errors when downloading bootstrap JS, it's doing that on other PRs, too.

[SherazKhan]

Thanks, @larsoner, and @dengemann. I think we need to discuss this in detail.