numpy.einsum error: too many subscripts for operand

Question

I am trying to use einsum to perform tensor multiplication. I am working in MATLAB, but am using the python interface to call numpy.einsum as described in this Q&A. Below is the code I'm using to perform the multiplication:

np = py.importlib.import_module('numpy');
ndim = 3; i = eye(ndim);
I = np.einsum('il,jk',matpy.mat2nparray(i),matpy.mat2nparray(i));
I4 = matpy.nparray2mat(...
  np.einsum('ijxyz,klxyz -> ijklxyz', I, matpy.mat2nparray(ones([30,30,30])) ));

However, the above code fails with the following error:

ValueError: einstein sum subscripts string contains too many subscripts for operand 0

Could you please provide a reproducible example ([mcve])? What's `ndim`? — Dev-iL, May 13 '19 at 12:29
ndim is just a arbitrary number, I do the calculation in 3D space, so ndim there is 3. — fenfam, May 14 '19 at 14:12
In the second call to `einsum` the first array is 4 dimensional, the second is 3, yet you're using 5 indices in the einsum call for both. That's the problem. If you just want to take a kind of outer product, perhaps you're looking for `'ijkl,xyz -> ijklxyz'`? Though this probably doesn't need einsum. — Andras Deak -- Слава Україні, May 14 '19 at 14:34
If my hunch is correct then `I = i[..., None, None] * i` and then `I4 = I[..., None, None, None] * np.ones((30, 30, 30))` to broadcast. — Andras Deak -- Слава Україні, May 14 '19 at 14:39
I'm told you can't use `None` to inject singleton dimensions when calling python from MATLAB. You can still use `reshape`, but I guess you can just sick with `einsum` in that case. — Andras Deak -- Слава Україні, May 14 '19 at 14:48

numpy.einsum error: too many subscripts for operand

0 Answers0