Unexpected behaviour in ragged tensor shapes

Question

I Have a list of numpy arrays, X, where the first dimension is ragged, but the others are of fixed length,

>> [x.shape for x in X] 
[(40, 2