By default we have average=False in compute_proj_ecg and compute_proj_eog. But when I was toying with the idea of using Xdawn (which I think might still be a good idea) to compute projectors rather than plain SSP, I found that average=True seemed to always be better than average=False.

Here is the sample data, with four columns: the first two are average=False, the last two are average=True; within those, the left are ECG epochs (to be projected) and the right are auditory epochs (hopefully not contaminated):

Note that the "average=True" does not really make sense for Xdawn because it already uses the average, so the first two and last two columns are the same for Xdawn.

I think we might consider:

• 1. Changing the default average=False to average=True.
• 2. Probably unify compute_exg_proj and create_exg_epochs. The former does not use the latter but it probably should.
• 3. Implement Xdawn as a backend instead of just SSP. Even with average=True it should do better against other noise present in the epochs, especially if that noise is large (yet to be shown).