Convert a 1D tensor to 2D tensor

[russellb23]

Using numpy we can convert 1D array to 2D array as,

a=np.arange(1,10)
a.reshape((3,-1))
array([[1, 2, 3],
       [4, 5, 6],
       [7, 8, 9]])

I am a newbie arraymancer user. Is there any way to convert 1D to 2D like the above?

[zetashift]

Are your perhapse looking for reshape like used here: https://mratsim.github.io/Arraymancer/tuto.shapeshifting.html ? let a = toSeq(1..9).toTensor().reshape(3,3)

EDIT and links to the doc: https://mratsim.github.io/Arraymancer/shapeshifting.html#reshape%2CTensor%2Cvarargs%5Bint%5D

[russellb23]

Thanks @zetashift . However with numpy -1 indicates each row to be added as a new axis essentially creating 3 x 3 2D array. With the numpy syntax, it is only enough to specify just number of rows and the columns will be computed automatically with -1 which is newaxis in numpy. I was wondering is there any such trick available which I am not aware of or couldn't find.

[mratsim]

No Arraymancer doesn't support that at the moment. It feels a bit like magic. -1 can mean a lot of things in the Python world so I'd rather not re-use -1 especially.

In terms of use-case, is that used often?

[DylanModesitt]

Aren't .squeeze and .unsqueeze good enough if you want implicit expansion?

[zerothi]

No Arraymancer doesn't support that at the moment. It feels a bit like magic. -1 can mean a lot of things in the Python world so I'd rather not re-use -1 especially.

In terms of use-case, is that used often?

Yes, using - 1 is really used a lot. For arrays where one of the dimensions is variable, then it would be nice with a fill shape.