Log accuracy score in EEGClassifier

[PierreGtch]

closes #510 closes #521

[codecov]

Codecov Report

Merging #541 (c62811b) into master (cb5cad0) will increase coverage by 0.19%.
Report is 1 commits behind head on master.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #541      +/-   ##
==========================================
+ Coverage   84.55%   84.75%   +0.19%     
==========================================
  Files          63       63              
  Lines        4676     4749      +73     
==========================================
+ Hits         3954     4025      +71     
- Misses        722      724       +2     

[bruAristimunha]

is there somewhere missing?

[PierreGtch]

@bruAristimunha all good to merge

[bruAristimunha]

Thank you @PierreGtch =)

[robintibor]

I noticed the default valid accuracy callback now causes me quite some issues in writing examples, things are logged twice etc. Was there a specific reason this was needed? @PierreGtch Considering to undo this part (not the cross entropy loss default).

[PierreGtch]

@robintibor I just thought it made sense to log the accuracy by default for classifiers.
If it helps, see how I deactivated it in the plot_sleep_staging_usleep.py example: https://github.com/braindecode/braindecode/pull/541/files#diff-6fce3d4fb0b5aed7465d731b19bbf3ff2a3016d32e1d6e58d006fb5bf67a3b74
Otherwise, if it creates too many issues, we can remove it.