Geositta2000

Slow contractions when scalars are present and output is large

3

The question in this issue is, suppose I have a contraction involving a scalar, `\sum_b 0.5 A[a,b] B[b,c] = C[a,c]`, in principle, the timing will be similar to `\sum_b A[a,b]...

Empty step in contractions with scalar by dp

1

For the following code ``` from opt_einsum import contract import numpy as np import opt_einsum as oe import timeit dim = 30 I = np.random.rand(dim, dim, dim, dim) C =...

Zero dimension tensors (as zero entries) lead opt_einsum slower than einsum

1

For contractions between tensors with various dimensions, sometimes the dimension may be zero, and I regarded it as a zero entry. Seems this is also what `einsum` does (refer to...

Contigugous loss after opt-einsum that may affect the efficiency of subsequent operations

8

After a tensor contraction, the resulting tensor from `numpy.einsum` is C-contiguous, and in `opt_einsum`, such properties may be lost. If there is a further tensor addition, the non-contiguous array may...

Hi, I tried to use the python API of `hptt` in analogy to `numpy`. For the following code, I met "ValueError: repeated axis in transpose" in `hptt.ascontiguousarray`, not in `np.ascontiguousarray`....