rl.dqn.DQN Class Reference

The DQN class. More...

Public Member Functions
def	__init__ (self, int number_of_actions, int state_size, network_path="")
	The RobotNetwork class initializer. More...
def	save (self, str path)
	Save the network to a given path. More...
def	update (self, np.ndarray state, np.ndarray next_state, int action, float reward, bool finished)
	Update the network, should happen between each state transition during training. More...
int	get_best_action (self, np.ndarray state)
	Look up the best action for a given state. More...

Public Attributes
	eps
	gamma
	eps_end
	eps_decay
	batch_size
	mem_size
	hidden_nodes
	network
	optimizer
	memory

Detailed Description

The DQN class.

Defines the class that will represent the Deep Q-Network.

Constructor & Destructor Documentation

__init__()

def rl.dqn.DQN.__init__	(		self,
		int	number_of_actions,
		int	state_size,
			network_path = ""
	)

The RobotNetwork class initializer.

Parameters

number_of_actions	The number of hidden nodes for the hidden layers.
state_size	The number of variables in a state, important for the nodes needed as input.
network_path	The path to the file that contains a (trained) network.

Returns

An instance of the DQN class.

Member Function Documentation

get_best_action()

int rl.dqn.DQN.get_best_action	(		self,
		np.ndarray	state
	)

Look up the best action for a given state.

Parameters

state

An input state representing the environment.

Returns

An action, represented as a number.

save()

def rl.dqn.DQN.save	(		self,
		str	path
	)

Save the network to a given path.

Parameters

path	The file path to save the network to.

update()

def rl.dqn.DQN.update	(		self,
		np.ndarray	state,
		np.ndarray	next_state,
		int	action,
		float	reward,
		bool	finished
	)

Update the network, should happen between each state transition during training.

Parameters

state	The start state of the environment.
next_state	The resulting, next state after executing an action.
action	The action that was executed.
reward	The reward resulting from the action.
finished	If the next_state is a finishing state.

---

The documentation for this class was generated from the following file:

• src/rl/dqn.py