shape error for OpticalSeries.data

[bendichter]

(copy/pasted from email conversation with Ueli Rutishauser (@urut). Moved here for posterity)

Trying to upload the dataset you have been helping me with to DANDI. i fixed all the things you sent and also other stuff that 'dani validate' spit out, down to one error that i cannot figure out.

Its this: "Could not construct OpticalSeries object due to: OpticalSeries.__init__: incorrect shape for 'data' (got '(3,)', expected '([None, None, None], [None, None, None, 3])')")"

I'm at a loss of what is expected. This field is empty for me as it is external files. The matlab export command is:

stimulus = types.core.OpticalSeries('data',[0,0,0]', 'timestamps',  [start_times'], 'orientation', 'lower left', ...
      'format', 'external', 'distance', 0.7, 'field_of_view', [0.3, 0.4, 0.7], 'dimension', [1280,720, 3], ...
      'data_unit', 'meters', 'external_file', extFileList, 'external_file_starting_frame', [0]);

Tried various things [0,0,0], [0,0,0]', [nan,nan,nan], [nan] ... error never changes.

Any thoughts how to fix this appreciated!

Ueli

Ueli,

Yes, this is a sort of obscure way that we annotate shape constraints. If you create a dataset like this:

data = [[NaN, NaN]; [NaN, NaN]]
data(:,:,2) = [[NaN, NaN]; [NaN, NaN]]

that will create a 3-dimensional array that should pass this check.

Explanation, if you are curious:
See our shape constraint documentation here. This particular shape constraint specification is ([None, None, None], [None, None, None, 3]).This is a list of two allowable specifications, [None, None, None] or [None, None, None, 3]. The length of each specification specifies the dimensionality of the array. A number is a fixed dimension length, and None means any length. The first allowable shape is any 3-dimensional array, and is intended for a sequence of grayscale images of shape [time * x * y]. The second specification is for a 4-D array with the last dimension as 3, which would be a sequence of color images of shape (t * x * y * (r,g,b)).

Now that I see this, I see that this would be pretty impossible for someone to decipher, so we definitely need to fix that.