Transformation graphs about navis

[schlegelp]

If I understand you correctly then this this functionality already exists. A couple notes:

1. navis.transforms.registry.find_bridging_path(source, target) (an instance of navis.transforms.templates.TemplateRegistry) uses a graph to collect on-the-fly all required transforms to go from a given source to a given target even if that involves more than a single transform into a TransformSequence. This also works with inverse directions.
2. The above takes weights into account which can be used to prioritise "good" over "bad" transforms (although I'm not making use of that at the moment).
3. As you pointed out in certain cases it might be beneficial to perform a certain operation in a specific brain space. Mirror is the first thing that comes to mind and indeed, navis.mirror_brain already has a via parameter that when set will transform the data to space X, perform the mirroring and then transform back to the original space.
4. Currently, navis only knows two types of transforms: "bridging" to go from one space to the other, and "mirror" to offset left/right asymmetries. We could add more types but so far I have resorted to just making up new templates spaces as required. For example to move between larval T1 and T2 segments, I would register a transform between "T1" and "T2" and then use navis.xform(some_data, source='T1', target='T2').

Does that already cover what you had in mind? I'd be happy to extend the system as required (as long as it remains reasonably simple).

[schlegelp]

Just to note that I copied that design pretty much 1:1 from Greg's nat.

[clbarnes]

Yes, that's it! Sorry I didn't find it :s

[schlegelp]

Good idea though! Keep'em coming :)