Fix export to EDF format with a set physical range smaller than the data range

[mscheltienne]

Today, I had a researcher from our sleep lab that came-in with a ~8 hours long night recording that could not be converted to EDF either through the BrainVision software or through MNE. The converted signals were all "flat" (or mostly flat), and this was due to 10' in the recording (around hour 7 ) where the signal went completely off-scale, for whatever reason.

The physical range is estimated here

mne-python/mne/export/_edf.py

Lines 122 to 131 in 6384a89

	if physical_range == 'auto':
	# get max and min for each channel type data
	ch_types_phys_max = dict()
	ch_types_phys_min = dict()
	for _type in np.unique(ch_types):
	_picks = np.nonzero(ch_types == _type)[0]
	_data = raw.get_data(units=units, picks=_picks)
	ch_types_phys_max[_type] = _data.max()
	ch_types_phys_min[_type] = _data.min()

On a good file, I get: pmin = ~ -130k and pmax= ~ 80k. The range is in the ~ 250k scale. On the bad file, the range was in the ~ 10M or 100M range, several order of magnitude larger. Thus, the conversion failed.

By setting the physical_range to (-150000, 80000) in the BrainVision software or in MNE, the data could be exported to EDF and the 10' segment off-scale is saturating as expected. However, in MNE, I did have to comment out this sanity-check

mne-python/mne/export/_edf.py

Lines 143 to 149 in 6384a89

	# check that physical min and max is not exceeded
	if data.max() > pmax:
	raise RuntimeError(f'The maximum μV of the data {data.max()} is '
	f'more than the physical max passed in {pmax}.')
	if data.min() < pmin:
	raise RuntimeError(f'The minimum μV of the data {data.min()} is '
	f'less than the physical min passed in {pmin}.')

I am not familiar at all with the EDF format, and might be missing something; but IMO it should be a warning instead of a raise.

[larsoner]

Nice @mscheltienne , I can't find anything to complain about! In it goes