Worker.A3C: Defines an Agent based on the A3C-Algorithm

In MartinKies/USLR: Reinforcement Learning with R

Description

Internal function which is called from Train.A3C.

Usage

Worker.A3C(worker.number, game.state, model.par, algo.par, algo.var.worker,
  game.object, weights, Worker.Episode, model)

Arguments

worker.number	Number to identify the given worker.
game.state	Game State from the last round.
model.par	A list as given by Get.Def.Par.Neural.Network.A3C or Get.Def.Par.Neural.Network.A3C.LSTM.
algo.par	A list as given by Get.Def.Par.A3C.
algo.var.worker	A list as given by Initialise.A3C.
game.object	A Game Object (list) as defined by Get.Game.Object.<NAME>.
weights	Network weights from the Master Network.
Worker.Episode	The actual Episode of the Worker.
model	A Model as given by Setup.Neural.Network.A3C or Setup.Neural.Network.A3C.LSTM

Details

Returns a list with the following items

• training A list containing calculated gradients.

• reward Last obtained reward during the current n-steps.

• finished Boolean specifiying wether episode is finished or not.

• algo.var.Worker List containing the stored experience of the given worker.

MartinKies/USLR documentation built on Nov. 10, 2019, 5:24 a.m.