Proposal: Add analyzer.frame_slice

[chrishalcrow]

Following on from #4133, here is a draft proposal for implementing a frame_slice method for the sorting analyzer, to see how tricky it is.

API is:

sliced_analyzer = analyzer.frame_slice(start_frame=1000, end_frame=2000)

which matches the frame_slice methods for sorting and recording objects.

There's a "hard" mode (which just recomputes all extensions) and a "soft" mode (which attempts to estimate the extensions). Almost all the complexity occurs at the level of the extension. For each extension, we'd need to implement a _frame_slice_extension_data method.

I've only implemented _frame_slice_extension_data for random_spikes, and you can only return an in memory analyzer for the moment.

Would like some feedback on the basic structure before doing anything more.

[yger]

I think this is a great idea, and I don't think of any obvisous flaws in the design as it is proposed. I'm afraid lots of extensions would need to be recomputed however, but this would be a cool feature

[alejoe91]

I agree with @yger here! It doesn't hurt and it could be beneficial in many cases! Of course, we should add proper warnings if things aren't as expected, e.g., too few sampled spikes, etc.