gyx

v0.1.25

gyx v0.1.25 Gyx.Agents.SARSA

This agent implements SARSA, it takes into account the current state, action, reward (st, at, rt) and on policy estimates for the best next action at+1 and state st+1.


The Q update is given by:

sarsa


The Q table process must be referenced on struct Q key, which must follow the Gyx.Qstorage behaviour

Link to this section Summary

Types

t()

Functions

act_epsilon_greedy(agent, environment_state)
act_greedy(agent, environment_state)
handle_call(arg, from, state)
init(process_q)
start_link(opts)
start_link(process_q, opts)
td_learn(agent, sarsa)

Link to this section Types

Link to this type

t() 
t() :: %Gyx.Agents.SARSA{
  Q: any(),
  epsilon: float(),
  epsilon_min: float(),
  gamma: float(),
  learning_rate: float()
}

Link to this section Functions

Link to this function

act_epsilon_greedy(agent, environment_state)

Link to this function

act_greedy(agent, environment_state)

Link to this function

handle_call(arg, from, state)

Link to this function

init(process_q)

Link to this function

start_link(opts)

Link to this function

start_link(process_q, opts)

Link to this function

td_learn(agent, sarsa)