how to reshape a ndarray? about ndarray HOT 6 OPEN

It doesn't have reshape() exactly, but you might be able to use the constructor ndarray(data[, shape, stride, offset]):

a = ndarray([1, 2, 3, 4], [2, 2])

b = ndarray(a.data, [4])

This will probably work perfectly well for simple cases where the array is not already a stepped or transposed view, but be advised that stride and offset are capable of representing memory layouts which may not be completely straightforward to reinterpret with a different shape. When I know what the input l looks like though, I've frequently used this approach.

TBH it does seem like a bit of missing functionality to do this with strict checking so that you know the result is valid.

from ndarray.

That's what I need, thanks.
However, after b = ndarray(a.data, [4]), it seems that b.data and a.data are bind together. When I change b.data, a.data changes as well.
Try this:

var a=zeros([1],'float32');
var b=ndarray(a.data,a.shape);
b.set(0,1);
console.log(a);
console.log(b);

you'll see that both a.data and b.data = [1]. That's weird.

from ndarray.

Yes, you can always call .slice() on the data array if you want to decouple them. You'll find that, in general, copying memory is the exception rather than the norm when using scijs/ndarray. This differs from matlab or numpy which tend to copy the memory on every operation. Also, in case you haven't seen it, ndarray-ops may be a useful set of functions, and I wrote this many years ago as well: https://github.com/scijs/scijs-ndarray-for-matlab-users

from ndarray.

Yes, .slice() works for me either. Thank you very much for all these information.

from ndarray.

Btw, are there any better modules to let me perform these operations as smoothly as pytorch?

from ndarray.

The only other module I know of is stdlib/ndarray. The Stdlib project is aiming to be much closer to numpy ndarrays, both in terms of design and extensibility, but although it's a much more complete foundation, it doesn't yet have an ecosystem of tools around it. I'm not too familiar with tensorflow, but if your interest is in ML, those toolkits might have more to offer than scijs/ndarray.

from ndarray.