I want to start by thanking you for all the good work on this library.
I have a question regarding the triangular_solve bindings.
I see in the issue #1076 and its related PR #1077, that we moved from torch_triangular_solve to torch_linalg_solve_triangular in the distributions-multivariate_normal.R file. I think this was because of the following deprecation warning:
[W BatchLinearAlgebra.cpp:2197] Warning: torch.triangular_solve is deprecated in favor of torch.linalg.solve_triangularand will be removed in a future PyTorch release.
I wanted to know if it was in the future plans to make the torch_linalg_solve_triangular a public binding for the torch interface and add some documentations to it, since, for now, we can only access it privately with :::. The publicly available function torch_triangular_solve still gives warning (as of 0.12.0), so maybe we could update its documentation (man/torch_triangular_solve.Rd) to reflect that or move to the linalg implementation?
The following is an example if it ever helps:
a <- torch_randn(c(3, 3))$triu_()
b <- torch_randn(3, 4)
# The following gives a warning on first execution
torch_triangular_solve(b, a, upper=TRUE)[[1]]
# The following is not publicly exposed but doesn't give a warning
torch:::torch_linalg_solve_triangular(a, b, upper=TRUE)
# The two above statements return the same results
Hello,
I want to start by thanking you for all the good work on this library.
I have a question regarding the triangular_solve bindings. I see in the issue #1076 and its related PR #1077, that we moved from
torch_triangular_solve
totorch_linalg_solve_triangular
in thedistributions-multivariate_normal.R
file. I think this was because of the following deprecation warning:I wanted to know if it was in the future plans to make the
torch_linalg_solve_triangular
a public binding for the torch interface and add some documentations to it, since, for now, we can only access it privately with:::
. The publicly available functiontorch_triangular_solve
still gives warning (as of 0.12.0), so maybe we could update its documentation (man/torch_triangular_solve.Rd
) to reflect that or move to the linalg implementation?The following is an example if it ever helps:
Many thanks in advance for your assistance.