Save results to a user given output directory

[tdincer]

Description

The changes made in this PR allows cnmf.fit_file method 1) to save the results in a user given output directory and 2) to return resultant motion correction object, if specified with a flag. All the changes made are backward compatible.

The motivation behind the changes is to incorporate CaImAn into a DataJoint workflow. We realized that the flexibility to save the results into a user given output directory was also thought by the CaImAn developers. As a reference please see the following block of code with comments from the load_memmap function in the mmaping.py file:

if pathlib.Path(filename).suffix != '.mmap':
        logging.error(f"Unknown extension for file {filename}")
        raise ValueError(f'Unknown file extension for file {filename} (should be .mmap)')
    # Strip path components and use CAIMAN_DATA/example_movies
    # TODO: Eventually get the code to save these in a different dir

The changes made in this PR requires only the pathlib python package.

Type of change

• [x] New feature (non-breaking change which adds functionality)

Has your PR been tested?

This PR has successfully passed both

python caimanmanager.py test

and

python caimanmanager.py demotest

prior to submitting our pull request.

[tdincer]

Hi @pgunn - I was wondering if you had a chance to look at this PR, and if so, about your opinion.

[pgunn]

Hello, Sorry for not getting back to you on this sooner; I've had a set of changes to Caiman that address this kind of concern (I've been thinking about it for awhile) that are not yet default behaviour. The changes are controlled by two new variables, if you'd like to play around with them (they are not well tested, but may do exactly what you want .. maybe).

Try this:

export CAIMAN_NEW_TEMPFILE=true
export CAIMAN_TEMP=/job/specific/path

(replace that value of CAIMAN_TEMP to something valid on your filesystem)

You'd do this with the version of caiman that does not have your patches; it should achieve the goal of keeping the temporary files of separate jobs distinct from each other.

I've been meaning to make these changes the default (after a lot more extensive testing), but other projects got in the way; I still mean to at some point.

[kushalkolar]

@pgunn Does setting CAIMAN_TEMP set the mcorr output memmaps to be stored in the CAIMAN_TEMP dir? (I'm looking for a way to get around #950 )

Thanks!

[pgunn]

It should (provided CAIMAN_NEW_TEMPFILE is also set to be true)

[kushalkolar]

I just tried this and it doesn't affect the location of the created memmap for motion correction. From searching the codebase it appears that caiman.paths.fn_relocated (which is the only function I could find that uses CAIMAN_NEW_TEMPFILE), is only used for online motion correction, not the offline motion correction:

https://github.com/flatironinstitute/CaImAn/search?q=fn_relocated%28

[pgunn]

Ah, sorry about that. I'll have to improve on this.

[kushalkolar]

@pgunn no worries, I could try to implement this for offline motion correction and do a PR later this week.

[EricThomson]

Where are we with this PR?

[kushalkolar]

I did the PR for mcorr so it uses the env var: https://github.com/flatironinstitute/CaImAn/pull/978

For CNMF you can already save the hdf file to whereever you want.

Longterm we should have a better storage system, perhaps in v2. This is part of what mesmerize does :)

[pgunn]

I'm going to look at this again soon given that you're working on it; given that this is something we've been working on for awhile and we probably want roughly the same end (but not necessarily the same implementation), we'll have to figure out if/how to get this merged in (and whether to sync it with my own efforts or not).