HMM pre-processing fails if sampling rate is not a multiple of 51.2

[AKuederle]

The "naive" upsampling implemnted here does not work with samplingrates that are not a even multiple

gaitmap/gaitmap_mad/gaitmap_mad/stride_segmentation/hmm/_hmm_feature_transform.py

Lines 261 to 277 in 8510fe0

	def inverse_transform_state_sequence(self, state_sequence: np.ndarray, *, sampling_rate_hz: float) -> np.ndarray:
	"""Inverse transform a state sequence to the original sampling rate.
	Parameters
	----------
	state_sequence
	The state sequence to be transformed back to the original sampling rate.
	This is done by repeating each state for the number of samples it was downsampled to.
	sampling_rate_hz
	The sampling rate of the original data in Hz
	Returns
	-------
	The state sequence in the original sampling rate
	"""
	return np.repeat(state_sequence, sampling_rate_hz / self.sampling_rate_feature_space_hz)

We should use proper interpolation using "nearest"

@HaMora @Aksei Jelena wanted to fix this I think? Any updates? And if she does not have time, could one of you have a look at that?

[Aksei]

I could take care of it. I am not 100% on this topic; never used the HMMs so far. My naive approach would be to use the script.signal.resample functionality instead of the np.repeat function.

@AKuederle @HaMora what di you think?

[AKuederle]

Seems reasonable. Does resample support a "nearest" option? Otherwise interpolate might do the job

[Aksei]

What do you mean with "nearest"?
The green signal is the one that needs to be resampled, isn't it?

[AKuederle]

I mean as a resample method. Because we need discrete values. With normal resample you would get float/in-between values

[AKuederle]

And yes, the green signal is the one that needs resampling

[HaMora]

Sorry for the late reply,
I read your comments and followed your conversation. I think, as @AKuederle mentioned, if we do resampling, we might need to do some aligning to match the two signals. But I think if we give the length of the original signal instead of the len(downsampled) * target_sampling_rate/actual_sampling_rate
(sort of interpolation) might be effective.

[AKuederle]

Resolved by #62