Save/restore Workspace

[mbillingr]

Certain computations such as MVARICA can take a lot of time but do not need to be performed very often because their results can be reused. It might be a useful feature for users of the high-level API to be able to dump and restore the entire Workspace.

I can see two possible implementation approaches:

1. Find out why the Workspace class cannot be pickled and see if that's easy to fix. (I suspect the culprit is a lambda function stored somewhere...)
2. Only save certain arrays from within the Workspace (mixing/unmixing matrices, VAR coefficients, activations, ...) and make the user responsible for setting up a Workspace with the correct configuration before loading.

Any thoughts? What are the use cases?

[cbrnr]

If I understand correctly, it might be useful to pre-compute connectivity features for the whole data set, which can then be used in cross-validating different classifiers. I wasn't aware that this isn't possible with the workspace - I now wonder if this can be implemented with the functional API?

Regarding the implementation, I'd prefer option 1. We had a similar issue in mne-tools/mne-python#6252, where we defined a custom __new__ method.

[mbillingr]

I think there is no problem with the functional API because most functions return numpy arrays or collections of numpy arrays. These can be easily saved and loaded using numpy's save and load routines.

The Workspace is basically a class that manages intermediate results without exposing them to the user. To store the current Workspace state we would need to support pickle or expose a save/load API that internally handles the intermediate arrays. If possible I would prefer option 1, too.