Open adam-hartshorne opened 9 months ago
I think there is a bug in the Concatenated Op, where the wrong axes are being processed.
e.g. A of size (100,2), B of size (100,1)
in Cola, if you use the Concatenated op via
C = cola.ops.Concatenated(A, B, axis=0)
C will be of size (200,2)...which isn't correct.
and
C = cola.ops.Concatenated(A, B, axis=1)
will give an Assertion Error: Trying to concatenate matrices of different sizes [(100, 2), (100, 1)]
Sorry about the late response! Good catch, I think the Concatenated op may not be as well tested as the other matrices. After the ICML deadline I will have a look into this
I think there is a bug in the Concatenated Op, where the wrong axes are being processed.
e.g. A of size (100,2), B of size (100,1)
in Cola, if you use the Concatenated op via
C = cola.ops.Concatenated(A, B, axis=0)
C will be of size (200,2)...which isn't correct.
and
C = cola.ops.Concatenated(A, B, axis=1)
will give an Assertion Error: Trying to concatenate matrices of different sizes [(100, 2), (100, 1)]