Custom policy reference¶

void MDP::RegisterPolicies(DynaPlex::Erasure::PolicyRegistry<MDP> &registry) const¶

Registers custom policies for the Markov Decision Process (MDP). This function is essential for adding heuristic policies that can be used for decision-making within the MDP framework.

Parameters:¶

DynaPlex::Erasure::PolicyRegistry<MDP>& registry

The registry where policies are stored and managed.

Returns:¶

None.

Rtype:¶

void

Examples

// Exact implementation of RegisterPolicies
void MDP::RegisterPolicies(DynaPlex::Erasure::PolicyRegistry<MDP>& registry) const
{
    registry.Register<BaseStockPolicy>("base_stock",
        "Base-stock policy with parameter base_stock_level - default parameter is equal"
        " to the bound on system inventory discussed in Zipkin (2008)");
}

int64_t CustomPolicy::GetAction(const MDP::State &state) const¶

The GetAction function in the CustomPolicy class calculates and returns the action to be taken for a given state in the MDP. This is only relevant for those that want to run their own custom policy, e.g., a heuristic baseline.

Parameters:¶

const MDP::State& state

The current state of the MDP, from which the action is determined.

Returns:¶

The action determined to be most appropriate for the given state.

Rtype:¶

int64_t

Usage Example

int64_t BaseStockPolicy::GetAction(const MDP::State& state) const
          {
                  int64_t action = base_stock_level - state.total_inv;
                  //We maximize, so actually this is capped base-stock.
                  if (action > mdp->MaxOrderSize)
                  {
                          action = mdp->MaxOrderSize;
                  }
                  return action;
          }