makeAgent: Create Agent.

In markdumke/reinforceR: Reinforcement Learning

Create Agent.

Description

An agent consists of a policy and (optional) a value function representation and (optional) a learning algorithm.

Usage

makeAgent(policy, val.fun = NULL, algorithm = NULL,
  preprocess = identity, replay.memory = NULL, policy.args = list(),
  val.fun.args = list(), algorithm.args = list())

Arguments

policy	[character(1) | Policy] A policy. If you pass a string the policy will be created via makePolicy.
val.fun	[character(1) | ValueFunction] A value function representation. If you pass a string the value function will be created via makeValueFunction.
algorithm	[character(1) | Algorithm] An algorithm. If you pass a string the algorithm will be created via makeAlgorithm.
preprocess	[function] A function which preprocesses the state so that the agent can learn on this.
replay.memory	[ReplayMemory] Replay memory for experience replay created by makeReplayMemory.
policy.args	[list] Arguments passed on to args in makePolicy.
val.fun.args	[list] Arguments passed on to args in makeValueFunction.
algorithm.args	[list] Arguments passed on to args in makeAlgorithm.

Examples

agent = makeAgent("softmax", "table", "qlearning")

markdumke/reinforceR documentation built on Nov. 17, 2022, 12:53 a.m.