Alexander Novikov

Automatic operation fusion

1

For example, instead of writing a separate function `project_matmul` we can implement something like `project_matmul = fuse(lambda a, b: project(a, b), lambda b, c: matmul(b, c))` To do that lets...

Something like ``` @support_batches def matmul(a, b): ... matmul(a, b) matmul(batch_a, b) matmul(batch_a, batch_b) ```

Project matmul complexity

1

Change the order of lhs and rhs einsums to (1, 2, 0, 3), 'aikd,sabc,bije,scjkf->sdef' and (1, 2, 0, 3), 'cikf,sdef,bije,sajkd->sabc' and change the complexity to O(d R n r**2 ry...

Fill gaps (Riemannian and DMRG support of TENSORBOX and Tensorlab), add https://github.com/rballester/ttrecipes https://github.com/goroda/Compressed-Continuous-Computation/tree/master/examples

https://github.com/Bihaqo/t3f/blob/develop/t3f/autodiff_test.py#L78

Better API for TTDense

1

move to t3f.layers, make a version without keras dependencies, rename rows, cols => inputs, outputs

Compare the speed of project matmul with t3f.gradient(lambda x: t3f.quadratic_form(H, x, x), x) and use this version of it's not slower