Call a (Potentially Unexported) Torch Function
Source:R/call_torch_function.R
call_torch_function.RdThis function allows calling a function prefixed with torch_, including unexported
functions which could have potentially valuable uses but which do not yet have
a user-friendly R wrapper function. Therefore, this function should be used with
extreme caution. Make sure you understand what the function expects as input. It
may be helpful to read the torch source code for help with this, as well as
the documentation for the corresponding function in the Pytorch C++ API. Generally
for development and advanced use only.
Arguments
- name
Name of the function to call as a string. Should start with "torch_"
- ...
A list of arguments to pass to the function. Argument splicing with
!!!is supported.- quiet
If TRUE, suppress warnings with valuable information about the dangers of this function.
Examples
if (torch_is_installed()) {
## many unexported functions do 'backward' calculations (e.g. derivatives)
## These could be used as a part of custom autograd functions for example.
x <- torch_randn(10, requires_grad = TRUE)
y <- torch_tanh(x)
## calculate backwards gradient using standard torch method
y$backward(torch_ones_like(x))
x$grad
## we can get the same result by calling the unexported `torch_tanh_backward()`
## function. The first argument is 1 to setup the Jacobian-vector product.
## see https://pytorch.org/blog/overview-of-pytorch-autograd-engine/ for details.
call_torch_function("torch_tanh_backward", 1, y)
all.equal(call_torch_function("torch_tanh_backward", 1, y, quiet = TRUE), x$grad)
}
#> Warning: Because this function allows access to unexported functions, please use with caution, and
#> only if you are sure know what you are doing. Unexported functions will expect inputs that
#> are more C++-like than R-like. For example, they will expect all indexes to be 0-based instead
#> of 1-based. In addition unexported functions may be subject to removal from the API without
#> warning. Set quiet = TRUE to silence this warning.
#> [1] TRUE