[MRG] Too large windows

[gemeinl]

No description provided.

[codecov]

Codecov Report

Merging #304 (97b7687) into master (b65a1dc) will increase coverage by 0.02%.
The diff coverage is 85.71%.

@@            Coverage Diff             @@
##           master     #304      +/-   ##
==========================================
+ Coverage   80.19%   80.22%   +0.02%     
==========================================
  Files          49       49              
  Lines        2944     2948       +4     
==========================================
+ Hits         2361     2365       +4     
  Misses        583      583              

[gemeinl]

braindecode/braindecode/preprocessing/windowers.py

Lines 182 to 185 in 150294c

	# check if recordings are of different lengths
	lengths = np.array([ds.raw.n_times for ds in concat_ds.datasets])
	if (np.diff(lengths) != 0).any():
	warnings.warn('Recordings have different lengths, they will not be batch-able!')

were wrong and actually are still wrong. It does not matter if recordings/trials are of different length, unless the user wants to extract a single window per recording/trial. So the check is incomplete

[gemeinl]

braindecode/braindecode/preprocessing/windowers.py

Lines 487 to 490 in 150294c

	if window_size_samples is not None and window_stride_samples is not None and \
	drop_last_window is None:
	raise ValueError('drop_last_window must be set if window_size_samples &'
	' window_stride_samples are not set')

is also incorrect. Error message does not match check above.

[gemeinl]

@dcwil could you have another look at these checks?

[dcwil]

braindecode/braindecode/preprocessing/windowers.py

Lines 182 to 185 in 150294c

	# check if recordings are of different lengths
	lengths = np.array([ds.raw.n_times for ds in concat_ds.datasets])
	if (np.diff(lengths) != 0).any():
	warnings.warn('Recordings have different lengths, they will not be batch-able!')

were wrong and actually are still wrong. It does not matter if recordings/trials are of different length, unless the user wants to extract a single window per recording/trial. So the check is incomplete

Should this block should also be checking if the user wants to extract a single window per trial, ie window_stride_samples and window_size_samples are None?

[gemeinl]

@dcwil Yes, correct. The warning only has to be raised if user attempts to extract one window per trial (in this case recording) and they are of different lengths. (Stride should not matter at all.) Usually, if window size is specified, or if trials are of the same length, all windows will always be of same length and therefore batch-able.

For the other check I don't know. I just saw that the error message does not match the check. But since the check is in an if else, I don't know whether something else has to be changed.

[robintibor]

Great that we caught this bug