Trait DiscreteActionMapper 

pub trait DiscreteActionMapper<A> {
    // Required methods
    fn num_actions(&self) -> usize;
    fn action_to_index(&self, action: &A) -> usize;
    fn index_to_action(&self, index: usize) -> A;
}

Maps between environment action types and integer indices.

DQN is a discrete-action algorithm. The Q-network outputs one Q-value per action, indexed 0..N. This trait bridges that integer world and the environment’s Action type, which may be an enum or something richer.

Example

use ember_rl::encoding::DiscreteActionMapper;
// CartPole: action is just usize (push left = 0, push right = 1)
struct CartPoleActions;
impl DiscreteActionMapper<usize> for CartPoleActions {
    fn num_actions(&self) -> usize { 2 }
    fn action_to_index(&self, action: &usize) -> usize { *action }
    fn index_to_action(&self, index: usize) -> usize { index }
}

Required Methods

fn num_actions(&self) -> usize

Total number of discrete actions available.

Determines the Q-network’s output layer size.

fn action_to_index(&self, action: &A) -> usize

Convert an action to its integer index.

Used when storing experience – we record the index, not the action.

fn index_to_action(&self, index: usize) -> A

Convert an integer index to an action.

Used when the Q-network selects an action – it returns an argmax index that we convert back to the environment’s action type.

Implementors

impl DiscreteActionMapper<usize> for UsizeActionMapper