resample() converts dtype to int16 even when dtype=‘uint16’ is specified

[carmengasco]

Hello!

When I load a binary file using read_binary, I specify dtype='uint16'. As expected, the trace I get is indeed of type uint16.

However, the issue arises when I use .resample(recording, 2500, dtype='uint16'). Even though I explicitly set the output dtype, the resulting data is automatically converted to int16.

Is there a way to preserve the original uint16 dtype during resampling?

Thanks a lot!

[h-mayorquin]

Which method is resample? Can you share a script that reproduces your error?

[carmengasco]

This is the resample method:
https://spikeinterface.readthedocs.io/en/stable/api.html#spikeinterface.preprocessing.resample

[zm711]

It seems like we have a note in the code that we need to force uint -> int

spikeinterface/src/spikeinterface/preprocessing/filter.py

Lines 406 to 410 in 604c049

	# if uint --> force int
	if dtype.kind == "u":
	dtype = np.dtype(dtype.str.replace("u", "i"))
	return dtype

Not sure why though.

[h-mayorquin]

I think this is for @alejoe91 and @samuelgarcia

[h-mayorquin]

The reason for having this behavior is that some scipy functions fail silently when taking uint as an input. So we have this all over preprocessing. I think that @alejoe91 recommended decimate as an alternative preprocessing method here but can't remember well.