multi indexing of functions

[mehrhardt]

I guess this issue was raised somewhere already but I could not find it. So please apologies my duplicate.

In all my applications, the images are 2 or 3 dimensional so it would be good if ODL was supporting multi indexing. Here an example that does not work:

>>> X = odl.uniform_discr(min_pt=[-1, -1], max_pt=[1, 1], shape=[128, 128])
>>> x = X.one()
>>> x[1,1]

Traceback (most recent call last):

  File "<ipython-input-13-3e593ebbc211>", line 1, in <module>
    x[1,1]

  File "odl/discr/discretization.py", line 326, in __getitem__
    return self.ntuple.__getitem__(indices)

  File "odl/space/npy_ntuples.py", line 367, in __getitem__
    arr = self.data[indices]

IndexError: too many indices for array

Of course by now I know a way around it: x.asarray()[1,1] However, this does not satisfy my needs as I want to write code that I can interchangeably with ODL or numpy. Is it so difficult to implement this directly, e.g. x[1,1] := x.asarray()[1,1]?

[adler-j]

I guess this issue was raised somewhere already but I could not find it. So please apologies my duplicate.

There is a huge effort in this direction, started with issue #342, then continued in #731, #857, #907, #908. The pull requests #739 and #861 are our WIP implementations of this.

Of course by now I know a way around it: x.asarray()[1,1] However, this does not satisfy my needs as I want to write code that I can interchangeably with ODL or numpy. Is it so difficult to implement this directly, e.g. x[1,1] := x.asarray()[1,1]?

We agree and we're working on it. However, on a deeper level this is non-trivial since we want to be able to support several backends, and up until lately there were no good GPU backends that allowed that syntax (there now is in libgpuarray).

You can anyway expect this feature to come to ODL Soon®.

[adler-j]

Can I close this as duplicate btw, or is there anything new from the others this adds?

[kohr-h]

Just for the record, I'm pushing this stuff for finalization currently, including GPU support. So hang tight :-)

[kohr-h]

Closing as duplicate of ~10 issues taken together.