Sessions Attribute Functionality Discussion.

[ljchang]

Working on sessions attribute for Fex objects. Right now users can pass in a sessions array that can be iterated over using Fex.itersessions(), exactly the same way that pd.DataFrame.iterrows() works.

Just wanted to run a few things by @jcheong0428 and @Nathaniel-Haines. If Fex.sesssions is not None, should we iterate over every unique session for all preprocessing/descriptive/feature extraction methods? For example, Fex.downsample() should probably downsample separately for each unique session (e.g., trial, subject) right? And Fex.clean() should do the same as well as Fex.extract_boft()

[ljchang]

Also, we need to make sure things like append() work sensibly. I assume we will want to append the session variables just like features.

We also might want to create a new session id when appending. Will probably add a flag for this too. Something like Fex.append(new_session=True)

[ljchang]

Do you think it is worth adding a ignore_sessions=False flag to all methods that we are autodetecting sessions?

[Nathaniel-Haines]

I agree that the default behavior should be to iterate over all unique session info. For the ignore_sessions=False flag, would it be better to have ignore_sessions accept a list of column headers? That way, if I want to ignore one session variable (e.g. subject) but not another (e.g. trial), I could set ignore_sessions=['subject']. This could be data added to the class to ensure it is inherited as we chain the methods.

[ljchang]

That won’t work as there can only be one session variable and I don’t think it would be a good idea to have more. If you want to switch session variables I think you should manually change it (eg from subject to trial). It’s too difficult to add in splitting by session variable otherwise. Right now it is a generator.

…

-luke (Sent from phone)

On Feb 11, 2018, at 10:53 AM, Nathaniel Haines ***@***.***> wrote: I agree that the default behavior should be to iterate over all unique session info. For the ignore_sessions=False flag, would it be better to have ignore_sessions accept a list of column headers? That way, if I want to ignore one session variable (e.g. subject) but not another (e.g. trial), I could set ignore_sessions=['subject']. This could be data added to the class to ensure it is inherited as we chain the methods. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[jcheong0428]

So I think two use cases should be considered here.
The first case is when processing just 1 Subject over multiple Trials.
This is easy with how sessions is currently coded as the user just needs to pass a a list of Trials corresponding to each timepoint.

However the user needs to do a little more in use case 2 which is when they first append over multiple Subjects x Trials, and try to batch processing on all of them (I think this is the case @Nathaniel-Haines is referring to). This would make it necessary to have a unique session over subject x Trial. Then the user would then need to first generate unique session values for those interactions then pass it into sessions. This shouldn't be too hard and we can provide a function for this.

I like the way @ljchang has it coded now for simplicity but we should definitely write in the doc about how to handle the second case (which would be common for trial-based experiments).

Also having all functions iterate over sessions should be clarified in docs due to different use cases. For instance baseline() or decompose() is probably better when the session=subject but not when session=trial. But once again this could be left to the users to check our doc for best practices.

[ljchang]

I think the current design could accomodate subject x trial, just done separately. First, user would process each subject by trial. Second, user would combine subject data and session now becomes subjects. Alternatively, users could come up with a unique subject x trial key (e.g., s001_t1), and process everything. I don't think it will be a good idea to combine subjects by trial in the sessions. We should be able to parallelize processes that iterate over sessions.

[jcheong0428]

Yup I agree managing unique sessions should be left to the user.

[ljchang]

Here is example for how to use sessions for a feature extraction method.

        if ignore_sessions:
            convolved = self.__class__(pd.DataFrame({x:convolve(y, wav, mode='same') for x,y in self.iteritems()}), sampling_freq=self.sampling_freq, features=self.features)
        else:
            convolved = self.__class__(sampling_freq=self.sampling_freq)
            for k,v in self.itersessions():
                session = self.__class__(pd.DataFrame({x:convolve(y, wav, mode='same') for x,y in v.iteritems()}), sampling_freq=self.sampling_freq, features=self.features)
                convolved = convolved.append(session, session_id=k)

We could also consider using groupby, but I think this will be easier in the long run.