`get_metadata_schema` in BaseImagingExtractorInterface needs a small refactor

[weiglszonja]

Accessing the private attribute sampling frequency and casting to float in get_metadata_schema is not a proper place to have anymore, I think it should be removed. https://github.com/catalystneuro/nwb-conversion-tools/blob/07e958d7564eb12d00a5ff5fabe661677d023efa/src/nwb_conversion_tools/datainterfaces/ophys/baseimagingextractorinterface.py#L27-L29

@CodyCBakerPhD what do you think?

[CodyCBakerPhD]

It shouldn't be necessary and is definitely not a good thing to do - I think this was a quick hack to make the tests work back when the code was first imported over. If you can add some code to ROIExtractors (and tests) that 100% guarantee every sampling frequency from every extractor returns a float, then yes that line will no longer be needed here.

[weiglszonja]

@CodyCBakerPhD How would you do it? I was thinking to add a decorator (yeah, I know...) that checks the return value of the get_sampling_frequency function:

def check_get_sampling_frequency_return_dtype(func):
    """ A decorator to check the return value of a function is of float type. """

    @wraps(func)
    def wrapper(*args, **kwargs):
        return_value = func(*args, **kwargs)
        if not isinstance(return_value, float):
            return_value = float(return_value)
        return return_value

    return wrapper

[bendichter]

We are already using typehints:

https://github.com/catalystneuro/roiextractors/blob/2361597b8234e988e6d70e661b7b20428ff32c99/src/roiextractors/imagingextractor.py#L38-L40

I think we should just ensure that we are following our own typehints and make sure all extractors return the appropriate type in the implementation of those extractors.

[CodyCBakerPhD]

@CodyCBakerPhD How would you do it? I was thinking to add a decorator (yeah, I know...)

I would simply write a test on the ROIExtractors side to ensure the output of each extractor .get_sampling_frequency is a float type.

The conversion tools shouldn't do too much magic like this on their own, they should ideally just expect particular forms of inptus/outputs and not behave correctly if that is violated. (the standard implication of type hints)

I did look into Pydantic type stuff for this a while back but the huge downside to that is they are not compatible with numpy data typing, and complicated unions of things can throw it off.

[bendichter]

yes, well put, Cody. Magic like this is difficult to maintain and tests are much simpler and just as effective.

[h-mayorquin]

The only place in the pipeline where we need this was changed recently and now does a casting over there: https://github.com/catalystneuro/nwb-conversion-tools/blob/07e958d7564eb12d00a5ff5fabe661677d023efa/src/nwb_conversion_tools/tools/roiextractors/roiextractors.py#L110

I just tested removing the line in the metadata and it works as it is.

[h-mayorquin]

Concerning checking for sampling rate equal being a float in imaging extractor, check: https://github.com/catalystneuro/roiextractors/pull/164

[h-mayorquin]

This is removed now in #559. Should we close the issue, @weiglszonja or is there something missing?

[weiglszonja]

Yeah its fine, I’ll close it.