1 #ifndef STAN_MATH_PRIM_FUN_BINARY_LOG_LOSS_HPP
2 #define STAN_MATH_PRIM_FUN_BINARY_LOG_LOSS_HPP
3
4 #include <stan/math/prim/meta.hpp>
5 #include <stan/math/prim/fun/log.hpp>
6 #include <stan/math/prim/fun/log1m.hpp>
7 #include <stan/math/prim/functor/apply_scalar_binary.hpp>
8 #include <cmath>
9
10 namespace stan {
11 namespace math {
12
13 /**
14 * Returns the log loss function for binary classification
15 * with specified reference and response values.
16 *
17 * The log loss function for prediction \f$\hat{y} \in [0, 1]\f$
18 * given outcome \f$y \in \{ 0, 1 \}\f$ is
19 *
20 * \f$\mbox{logloss}(1, \hat{y}) = -\log \hat{y} \f$, and
21 *
22 * \f$\mbox{logloss}(0, \hat{y}) = -\log (1 - \hat{y}) \f$.
23 *
24 * @tparam T value type
25 * @param[in] y reference value, either 0 or 1
26 * @param[in] y_hat response value in [0, 1]
27 * @return Log loss for response given reference value
28 */
29 template <typename T, require_arithmetic_t<T>* = nullptr>
binary_log_loss(int y,const T & y_hat)30 inline T binary_log_loss(int y, const T& y_hat) {
31 using std::log;
32 return y ? -log(y_hat) : -log1m(y_hat);
33 }
34
35 /**
36 * Enables the vectorised application of the binary log loss function, when
37 * the first and/or second arguments are containers.
38 *
39 * @tparam T1 type of first input
40 * @tparam T2 type of second input
41 * @param a First input
42 * @param b Second input
43 * @return Binary log loss function applied to the two inputs.
44 */
45 template <typename T1, typename T2, require_any_container_t<T1, T2>* = nullptr,
46 require_not_var_matrix_t<T2>* = nullptr>
binary_log_loss(const T1 & a,const T2 & b)47 inline auto binary_log_loss(const T1& a, const T2& b) {
48 return apply_scalar_binary(a, b, [&](const auto& c, const auto& d) {
49 return binary_log_loss(c, d);
50 });
51 }
52
53 } // namespace math
54 } // namespace stan
55
56 #endif
57