0 votes
asked by (120 points)

I've been messing around a bit with ITensor for some ML/signal processing applications and I think the library is awesome! However, I'm not sure how to go about the following: I have two "matrix" tensors - say P and Q - whose rows represent samples from a signal and whose columns represent features of the signal. In code:

uint num_samp = 100;
auto idx_samp = Index(num_samp "samples");
auto idx_f1 = Index(2, "features1");
auto idx_f2 = Index(2, "features2");
auto P = randomITensor(idx_samp, idx_f1);
auto Q = randomITensor(idx_samp, idx_f2);

Now I have some "operator" U indexed by idx_f1 and idx_f2 and some other index idx_u. I want to contract each row of P and Q with U producing a new tensor indexed by idx_samp and idx_u. In pseudocode:

U = randomITensor(idx_f1, idx_f2, idx_u);
V = ITensor(idx_samp, idx_u);
for n in 1:num_samp
    V[idx_samp=n,:] = P[idx_samp=n,:]*U*Q[idx_samp=n,:]

But I'm not sure how to go about doing this in ITensor. Of course, I could just treat everything as a matrix and not worry about it, but there are some other steps where it's nice to have ITensor keep track of bookeeping on indices. I'm new both to ITensor and thinking about things as tensors so apologies if the question has a straightforward answer.

1 Answer

0 votes
answered by (14.1k points)

You could write the loops explicitly as you show.

An alternative is to write it in terms of delta tensors to constrain idx_samp of P and Q to be the same value, similar to this question here: http://www.itensor.org/support/1407/computing-the-partial-kronecker-product-of-two-itensors?show=1407#q1407

So for your application, it would be something like:

auto del = delta(idx_samp,prime(idx_samp),prime(idx_samp,2));
auto V = (P*U*prime(Q,idx_samp)*del).noPrime();

As is mentioned in the other post, this creates some unnecessary intermediate tensors so may not be the most efficient solution. Please let us know if either approach is not fast enough, it would be nice to have a more direct and efficient way of doing this in ITensor (but that kind of operation has not been very common so far in physics applications, so hasn't been a high priority to make special functionality for it).

Welcome to ITensor Support Q&A, where you can ask questions and receive answers from other members of the community.

Formatting Tips:
  • To format code, indent by four spaces
  • To format inline LaTeX, surround it by @@ on both sides
  • To format LaTeX on its own line, surround it by $$ above and below
  • For LaTeX, it may be necessary to backslash-escape underscore characters to obtain proper formatting. So for example writing \sum\_i to represent a sum over i.
If you cannot register due to firewall issues (e.g. you cannot see the capcha box) please email Miles Stoudenmire to ask for an account.

To report ITensor bugs, please use the issue tracker.