Question/Suggestion: Rethink / Adjust to suboptimal up-/downsampling in to_dense()

[adswa]

We have been looking further into the issue of the extra column during resampling (#358, #365) and ended up with a bit of confusions regarding the implementation of to_dense().

pybids/bids/variables/variables.py

Lines 326 to 341 in e9fc2df

	duration = int(math.ceil(sampling_rate * self.get_duration()))
	ts = np.zeros(duration, dtype=self.values.dtype)
	onsets = np.round(self.onset * sampling_rate).astype(int)
	durations = np.round(self.duration * sampling_rate).astype(int)
	run_i, start, last_ind = 0, 0, 0
	for i, val in enumerate(self.values.values):
	if onsets[i] < last_ind:
	start += self.run_info[run_i].duration * sampling_rate
	run_i += 1
	_onset = int(start + onsets[i])
	_offset = int(_onset + durations[i])
	ts[_onset:_offset] = val
	last_ind = onsets[i]

It seems to be that during up or downsampling, the data is simply "broadcasted" instead of interpolated? Downsampling via this broadcasting and subsequent linear interpolation during 'proper' downsampling in resample() would then change the data even if nothing was done to it. @yarikoptic suggest that the default kind in resample should be 'nearest' if its broadcasting instead of linear interpolation here. Or instead to use linear interpolation to fill the values.

Are we thinking correct here?

[tyarkoni]

Not sure I follow... to_dense can only be called on a SparseRunVariable. Each event in a SparseRunVariable is represented strictly in terms of onset, duration, and amplitude. We're just taking that temporal information and filling a uniformly-sampled array with boxcars that have the appropriate onsets and offsets. There's no resampling happening. (I think maybe your confusion is that both sparse and dense variables store their values in .values. But the meaning is different, and since to_dense can't be called on a DenseRunVariable, you would never end up doing any actual resampling of timeseries data.)

[yarikoptic]

Heh, I think it is my bad reading of the code here. Sorry about the noise

[tyarkoni]

No worries—happy to tolerate a little noise given all the signal today. :)