mirror of
https://github.com/HChaZZY/Stockfish.git
synced 2025-12-26 03:56:50 +08:00
Simplify passing constants that may vary between calls.
This commit is contained in:
@@ -209,7 +209,7 @@ namespace Learner::Autograd::UnivariateStatic
|
||||
struct VariableParameter : Evaluable<T, VariableParameter<T, I>>
|
||||
{
|
||||
using ValueType = T;
|
||||
|
||||
|
||||
static constexpr bool is_constant = false;
|
||||
|
||||
constexpr VariableParameter()
|
||||
@@ -281,6 +281,36 @@ namespace Learner::Autograd::UnivariateStatic
|
||||
T m_x;
|
||||
};
|
||||
|
||||
// The "constant" may change between executions, but is assumed to be
|
||||
// constant during a single evaluation.
|
||||
template <typename T>
|
||||
struct ConstantRef : Evaluable<T, ConstantRef<T>>
|
||||
{
|
||||
using ValueType = T;
|
||||
|
||||
static constexpr bool is_constant = true;
|
||||
|
||||
constexpr ConstantRef(const T& x) :
|
||||
m_x(x)
|
||||
{
|
||||
}
|
||||
|
||||
template <typename... ArgsTs>
|
||||
[[nodiscard]] T calculate_value(const std::tuple<ArgsTs...>&) const
|
||||
{
|
||||
return m_x;
|
||||
}
|
||||
|
||||
template <typename... ArgsTs>
|
||||
[[nodiscard]] T calculate_grad(const std::tuple<ArgsTs...>&) const
|
||||
{
|
||||
return T(0.0);
|
||||
}
|
||||
|
||||
private:
|
||||
const T& m_x;
|
||||
};
|
||||
|
||||
template <typename LhsT, typename RhsT, typename T = typename std::remove_reference_t<LhsT>::ValueType>
|
||||
struct Sum : Evaluable<T, Sum<LhsT, RhsT, T>>
|
||||
{
|
||||
|
||||
@@ -220,6 +220,16 @@ namespace Learner
|
||||
return loss_;
|
||||
}
|
||||
|
||||
template <typename ValueT>
|
||||
static auto& expected_perf_(ValueT&& v_)
|
||||
{
|
||||
using namespace Learner::Autograd::UnivariateStatic;
|
||||
|
||||
static thread_local auto perf_ = sigmoid(std::forward<ValueT>(v_) * ConstantRef<double>(winning_probability_coefficient));
|
||||
|
||||
return perf_;
|
||||
}
|
||||
|
||||
static ValueWithGrad<double> get_loss(Value shallow, Value teacher_signal, int result, int ply)
|
||||
{
|
||||
using namespace Learner::Autograd::UnivariateStatic;
|
||||
@@ -238,18 +248,17 @@ namespace Learner
|
||||
auto loss_ = pow(q_ - p_, 2.0) * (1.0 / (2400.0 * 2.0 * 600.0));
|
||||
*/
|
||||
|
||||
static thread_local auto q_ = sigmoid(VariableParameter<double, 0>{} * ConstantParameter<double, 4>{});
|
||||
static thread_local auto p_ = sigmoid(ConstantParameter<double, 1>{} * ConstantParameter<double, 4>{});
|
||||
static thread_local auto q_ = expected_perf_(VariableParameter<double, 0>{});
|
||||
static thread_local auto p_ = expected_perf_(ConstantParameter<double, 1>{});
|
||||
static thread_local auto t_ = (ConstantParameter<double, 2>{} + 1.0) * 0.5;
|
||||
static thread_local auto lambda_ = ConstantParameter<double, 3>{};
|
||||
static thread_local auto loss_ = cross_entropy_(q_, p_, t_, lambda_);
|
||||
|
||||
auto args = std::tuple(
|
||||
(double)shallow,
|
||||
(double)teacher_signal,
|
||||
(double)result,
|
||||
calculate_lambda(teacher_signal),
|
||||
winning_probability_coefficient
|
||||
(double)shallow,
|
||||
(double)teacher_signal,
|
||||
(double)result,
|
||||
calculate_lambda(teacher_signal)
|
||||
);
|
||||
|
||||
return loss_.eval(args).clamp_grad(max_grad);
|
||||
|
||||
Reference in New Issue
Block a user