ica.get_sources has strange ch_names and ch_type

[ppasler]

After calling ica.get_source() with a Raw object I noticed that

• the ch_names "ICA 001" start by 1 instead of 0 (which is not consistent)
• I couldn't plot it right, as the ch_type for scaling is not clear (it seems to be misc, but I couldn't find any documentation about it)

I would suggest to let the ICAs start by 0 everywhere.
Remove the +1 here: https://github.com/mne-tools/mne-python/blob/master/mne/preprocessing/ica.py#L798

[agramfort]

this question has been raised so many times... So far we agreed that when you read "ICA #1" it means index 1 but then names start with 1 to be consistent with standard EEG channels on Elekta : EEG001 etc.

do we really have to change this?

[ppasler]

Sorry, I did not check for older tickets.
You've got a point with consistency of EEG channel naming.

I just noticed, that inside the ICA object the ICs start with 0 (e.g. ica.plot_sources), but after transferring to a Raw object it starts with 1 - so I would propose to change this within the ICA object. It's a common computer science problem that indices start with 0...

The other thing is to add the hint 'the channel type is misc' when calling ica.get_source to the doc.

[agramfort]

channel names start with one and indices eg in exclude start at zero. Also ica plot methods refer to ICA indices and start at zero. +1 to clarify the doc

[jasmainak]

It seems to be a good candidate for the FAQ section :)

[mmagnuski]

I also think that ic numbering inconsistency is counterintuitive (and changing the docs won't change the intuitions ;) ).
ica.plot_sources plots sources with names in IC #000 convention - which is good as it makes comparisons between plot_components and plot_sources easier, but further increases surprise when ica.get_sources().plot() gives other labels (ICA 001). If channel labels must of form ICA 001 - then maybe ICA plotting functions should print ICA 001 instead of the index? Or maybe both should change to IC 001? 😈

[larsoner]

So over in #4320 the crushing weight of a half dozen (or more) opened issues/PRs about this issue has caused @agramfort to have a change of heart to accept the possibility of naming channels as ICA 000+. It creates inconsistency with other channel names, but IMO (and maybe @agramfort has come around to it, now, too) this is less bad / create fewer problems than our current solution, where most users are confused by our ICA interface.

@dengemann @jona-sassenhagen @jasmainak @mmagnuski are you okay with this change?

@mmagnuski would you perhaps be up for making a PR?

[jona-sassenhagen]

I'm in favour.

[dengemann]

I'll let you converge on this. After 10 MNE workshops I learned that user logic is sometimes as important as designer logic :) let's make it trivial and unsurprising to use ICA!

…

On Thu, 15 Jun 2017 at 15:53, jona-sassenhagen ***@***.***> wrote: I'm in favour. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3889 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AB0fighF33nQTnZUErDBn-8LCDhG6sjPks5sETdJgaJpZM4LenEh> .

[mmagnuski]

I'm also +1 on this. I'm travelling now, but I'll try to get to it this week and make a PR.
(BTW - is anyone attending cutting eeg?)

[larsoner]

Actually @mmagnuski I had a look and the past three attempts all do the same thing as the current attempt, except #1622 which had the option on how to name them, which we probably need. I'll see if it can be worked into #4320 because the repo for #1622 is gone.

[dengemann]

Maybe we can also make explicit in the label that it's an index.

…

On Thu, 15 Jun 2017 at 16:09, Eric Larson ***@***.***> wrote: Actually @mmagnuski <https://github.com/mmagnuski> I had a look and the past <#3633> three <#2019> attempts <#1622> all do the same thing as the current attempt <#4320>, except #1622 <#1622> which had the option on how to name them, which we probably need. I'll see if it can be worked into #4320 <#4320> because the repo for #1622 <#1622> is gone. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3889 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AB0fivkJCH-yxmZsB1l7eL7YJxl9tts9ks5sETpcgaJpZM4LenEh> .

[teonbrooks]

to add to the list of past issues: #2000 ;)

Probably said already but one of the most confusing parts has been the IC vsICA labelling. how do we deal with indexing channels because this problem doesn't come up? Two options:

• get rid of IC # completely and remain consistent in ICA label and use the ch indexing methods we use for other channel types.
• index with 0 in the name. used to be my preferred option but not such a strong advocate for/against it anymore

[larsoner]

I think this has been fixed in master via naming and/or doc updates