Open AlexandreBrown opened 1 year ago
In your example where a
is shaped (W, H)
, using createAlignedNDArray(a.toListD3(), 0.0)
will give the same shape (W, H). ndarray in multik is aligned, i.e. all sizes at dimension 0 will be the same. Unfortunately, it will not work to get the shape (P, P)
. This can be done by converting the output lists to a mutableList and expanding one of them to size P.
I had a request for this function, having figured it out, I called it createAlignedNDArray
, since this is not a full-fledged pad
.
In that case, I'll try to add the classic pad
.
I would like to know how to pad using
createAlignedNDArray
.From my understanding this function takes a list of lists and makes every list the max length and pads the remaining with zeros or whatever value you want (for each dim).
Example:
Let
P
be the size of each dims we want our final matrix to be with padding.Let
W
andH
be 2 sizes (different) that are less thanP
.For the sake of this example let's say we have
W
<H
<P
.Let
a
be an mk array of shape(W, H)
. Then applyingcreateAlignedNDArray(a.toListD3(), 0.0)
would give us a matrix of shape(H, H)
padded with zeros.While the name of the function seems correct, I'm interested in knowing if it's possible to use this function to transoform
a
into a matrix of size(P, P)
instead.Something like, we provide an mk array and can specify the desired padding for each dimensions.
See https://pytorch.org/docs/stable/generated/torch.nn.functional.pad.html
If it's possible with
createAlignedNDArray
then feel free to let me know.