Automatic Differentiation
 
Loading...
Searching...
No Matches
softmax.hpp
Go to the documentation of this file.
1#ifndef STAN_MATH_OPENCL_REV_SOFTMAX_HPP
2#define STAN_MATH_OPENCL_REV_SOFTMAX_HPP
3#ifdef STAN_OPENCL
4
11
12namespace stan {
13namespace math {
14
23template <typename T,
24 require_all_kernel_expressions_and_none_scalar_t<T>* = nullptr>
26 return make_callback_var(
27 softmax(A.val()), [A](vari_value<matrix_cl<double>>& res) mutable {
28 if (res.val().size() == 0) {
29 return;
30 }
31 A.adj() += elt_multiply(
32 res.val(), (res.adj() - dot_product(res.adj(), res.val())));
33 });
34}
35
36} // namespace math
37} // namespace stan
38
39#endif
40#endif
Represents an arithmetic matrix on the OpenCL device.
Definition matrix_cl.hpp:47
elt_multiply_< as_operation_cl_t< T_a >, as_operation_cl_t< T_b > > elt_multiply(T_a &&a, T_b &&b)
var_value< plain_type_t< T > > make_callback_var(T &&value, F &&functor)
Creates a new var initialized with a callback_vari with a given value and reverse-pass callback funct...
auto softmax(T &&x)
Return the softmax of each vector in a container of fvar values.
Definition softmax.hpp:23
auto dot_product(const T_a &a, const T_b &b)
Returns the dot product of the specified vectors.
The lgamma implementation in stan-math is based on either the reentrant safe lgamma_r implementation ...