Automatic Differentiation
 
Loading...
Searching...
No Matches
log_softmax.hpp
Go to the documentation of this file.
1#ifndef STAN_MATH_PRIM_FUN_LOG_SOFTMAX_HPP
2#define STAN_MATH_PRIM_FUN_LOG_SOFTMAX_HPP
3
10#include <utility>
11
12namespace stan {
13namespace math {
14
45template <typename Container, require_st_arithmetic<Container>* = nullptr,
46 require_container_t<Container>* = nullptr,
47 require_not_t<bool_constant<
48 is_eigen<std::decay_t<Container>>::value
49 && !is_eigen_vector<std::decay_t<Container>>::value>>* = nullptr>
50inline auto log_softmax(Container&& x) {
51 return make_holder(
52 [](auto&& a) {
54 std::forward<decltype(a)>(a),
55 [](auto&& v) -> plain_type_t<decltype(v)> {
56 if (v.size() == 0) {
57 return std::forward<decltype(v)>(v);
58 }
59 return (v.array() - log_sum_exp(v)).matrix();
60 });
61 },
62 to_ref(std::forward<Container>(x)));
63}
64
65} // namespace math
66} // namespace stan
67#endif
auto make_holder(F &&func, Args &&... args)
Calls given function with given arguments.
Definition holder.hpp:488
auto log_softmax(T &&x)
Return the log softmax of each vector in a container of fvar values.
ref_type_t< T && > to_ref(T &&a)
This evaluates expensive Eigen expressions.
Definition to_ref.hpp:18
constexpr decltype(auto) apply(F &&f, Tuple &&t, PreArgs &&... pre_args)
Definition apply.hpp:51
fvar< T > log_sum_exp(const fvar< T > &x1, const fvar< T > &x2)
typename plain_type< std::decay_t< T > >::type plain_type_t
The lgamma implementation in stan-math is based on either the reentrant safe lgamma_r implementation ...