Bad channel make it through the pipeline (or was caused by the pipeline).

[christian-oreilly]

This image is part of some diagnosis plot produced by @Deepa-Tilwani for our NSF Valence project (award #2419634):

In the clean PSD, there is one channel at less than -200 dB. There should be no such channel after applying the cleaning operation of PyLossless.

[christian-oreilly]

@Deepa-Tilwani Could you check what channel is having this aberrantly low PSD? Should be simple to get the PSD data through a function like psd_multitaper() and check which channel as a mean PDF across frequency below -100. Once we identified the channel, we can compare its time course before and after the pipeline along with looking at if it got flagged in some way by the pipeline. Hopefully the reason why it gets completely attenuated like this and the solution may become obvious. The only thing I can think of right now is if you include the reference channel as a channel... then this channel will become ref-ref ~= 0. Not sure how would that happen during the execution of the pipeline though.

[scott-huberty]

The only thing I can think of right now is if you include the reference channel as a channel... then this channel will become ref-ref ~= 0. Not sure how would that happen during the execution of the pipeline though.

That's where my thought went. I've run into plots similar to your "clean plot" When I compute and plot the PSD on an EEG file (and include the reference channel in the PSD.)

If I remember correctly, the pipeline does not be apply a reference (average or otherwise) to the input data.

[christian-oreilly]

Indeed, this channel is the reference (Cz) and the problem goes away if we set the average reference before plotting. But shouldn't the cleaned raw after applying the RejectionPolicy already have been set to an average reference rather than be set to a Cz reference? It seems weird to me that it did not.

[scott-huberty]

shouldn't the cleaned raw after applying the RejectionPolicy already have been set to an average reference rather than be set to a Cz reference?

No I don' think the RejectionPolicy sets a reference on the input data.

The pipeline sets it on the epochs that it internally generates but the input data doesn't get touched there either:

pylossless/pylossless/pipeline.py

Lines 752 to 754 in 9052a63

	if rereference and picks=="eeg":
	self.flags["ch"].rereference(epochs)