dqn_network.DeepQNetwork Class Reference

Public Member Functions
def	__init__ (self, list nums_actions, int num_features, learning_rate=0.01, reward_decay=0.9, e_greedy=0.9, replace_target_iter=300, memory_size=500, batch_size=32, e_greedy_increment=None, output_tensorboard=False, load_model_data=False)
def	build_output_net (self, w_initializer, b_initializer, c_names, l11_8, l12_8, l21_8, l22_8, n_l9)
def	build_sub_network (self, w_initializer, b_initializer, c_names, n_l1, n_l2, n_l3, n_l4, n_l5, n_l6, n_l7, n_l8, n_l9)
def	store_transition (self, observation, action, reward, observation_new)
def	choose_action (self, observation)
def	learn (self)
def	plot_cost_reward (self)
def	save_reward_loss (self, frequency=100)
def	store_network (self, model_name='MyModel', tar_path=os.path.dirname(__file__)+'/dqnNetModel/')
def	load_network (self, tar_path='/dqnNetModel/')

Private Member Functions
def	_build_net (self)

Static Private Member Functions
def	_create_dense (tf.Tensor input_tensor, list net_size, int start_layer_id, str layer_name, dict w_dir, dict b_dir)

Private Attributes
	_reward
	_n_actions
	_n_features
	_lr
	_gamma
	_epsilon_max
	_replace_target_iter
	_memory_size
	_batch_size
	_epsilon_increment
	_epsilon
	_output_tensorboard
	_learn_step_counter
	_memory
	_sess
	_replace_target_op
	_summary_writer
	_merged
	_cost11_his
	_cost12_his
	_cost21_his
	_cost22_his
	_reward_hit
	_memory_counter
	_s
	_q_target11
	_q_target12
	_q_target21
	_q_target22
	_q_eval22
	_loss11
	_loss12
	_loss21
	_loss22
	_train_op11
	_train_op12
	_train_op21
	_train_op22
	_s_
	_q_next22

Detailed Description

dqn 深度强化学习

Using:
    Tensorflow: 1.15

create:
    策略数，环境维度

choose_action:
    get observation, return action（0 - (num_actions-1))

store_transition: store data as memory
    get observation, action, reward, observation_new
    return none

learn: start one learn with memory
    get none, return none

plot_cost: print picture at the end
    get none, return none

get_reward_history
    get none, return reward list

get_cost_history
    get none, return cost list

store_network: save parameters at save path
    get save path, return none

load_network: load parameters at load path
    get load path, return none

Constructor & Destructor Documentation

__init__()

def dqn_network.DeepQNetwork.__init__	(		self,
		list	nums_actions,
		int	num_features,
			learning_rate = 0.01,
			reward_decay = 0.9,
			e_greedy = 0.9,
			replace_target_iter = 300,
			memory_size = 500,
			batch_size = 32,
			e_greedy_increment = None,
			output_tensorboard = False,
			load_model_data = False
	)

初始化
:param nums_actions: 列表，列表中每个数字表示各组别动作个数
:param num_features: 环境维度
:param learning_rate: 学习率
:param reward_decay: 奖励影响率（gamma）
:param e_greedy: 选择最优策略的概率（当直接使用训练好的模型时，可设置为1）
:param replace_target_iter: 隔多少拍更新一次tar网络参数
:param memory_size: 记忆池的容量
:param batch_size: 每次训练的组的数据大小
:param e_greedy_increment: 是否不断减小e_greedy
:param output_tensorboard: 是否输出tensorboard图
:param load_model_data: 是否直接从指定路径加载网络参数

Member Function Documentation

_build_net()

def dqn_network.DeepQNetwork._build_net (   self )

private

Build the neuron network 

_create_dense()

def dqn_network.DeepQNetwork._create_dense ( tf.Tensor  input_tensor, list  net_size, int  start_layer_id, str  layer_name, dict  w_dir, dict  b_dir  )

staticprivate

创建指定网络结构的CNN网络(默认各层都使用relu）
输出层需要另外建立
:param input_tensor: 网络输入, 要求为一个tensor:shape=[1, n]
:param net_size: 网络规格, 不包括输入层
:return: tensor

build_output_net()

def dqn_network.DeepQNetwork.build_output_net	(		self,
			w_initializer,
			b_initializer,
			c_names,
			l11_8,
			l12_8,
			l21_8,
			l22_8,
			n_l9
	)

build_sub_network()

def dqn_network.DeepQNetwork.build_sub_network	(		self,
			w_initializer,
			b_initializer,
			c_names,
			n_l1,
			n_l2,
			n_l3,
			n_l4,
			n_l5,
			n_l6,
			n_l7,
			n_l8,
			n_l9
	)

choose_action()

def dqn_network.DeepQNetwork.choose_action	(		self,
			observation
	)

Choose an action from observation 

learn()

def dqn_network.DeepQNetwork.learn

(

self

)

Start the learning process 

load_network()

def dqn_network.DeepQNetwork.load_network	(		self,
			tar_path = '/dqnNetModel/'
	)

plot_cost_reward()

def dqn_network.DeepQNetwork.plot_cost_reward

(

self

)

save_reward_loss()

def dqn_network.DeepQNetwork.save_reward_loss	(		self,
			frequency = 100
	)

store_network()

def dqn_network.DeepQNetwork.store_network	(		self,
			model_name = 'MyModel',
			tar_path = os.path.dirname(__file__) + '/dqnNetModel/'
	)

store_transition()

def dqn_network.DeepQNetwork.store_transition	(		self,
			observation,
			action,
			reward,
			observation_new
	)

Store the memory 

Member Data Documentation

_batch_size

dqn_network.DeepQNetwork._batch_size

private

_cost11_his

dqn_network.DeepQNetwork._cost11_his

private

_cost12_his

dqn_network.DeepQNetwork._cost12_his

private

_cost21_his

dqn_network.DeepQNetwork._cost21_his

private

_cost22_his

dqn_network.DeepQNetwork._cost22_his

private

_epsilon

dqn_network.DeepQNetwork._epsilon

private

_epsilon_increment

dqn_network.DeepQNetwork._epsilon_increment

private

_epsilon_max

dqn_network.DeepQNetwork._epsilon_max

private

_gamma

dqn_network.DeepQNetwork._gamma

private

_learn_step_counter

dqn_network.DeepQNetwork._learn_step_counter

private

_loss11

dqn_network.DeepQNetwork._loss11

private

_loss12

dqn_network.DeepQNetwork._loss12

private

_loss21

dqn_network.DeepQNetwork._loss21

private

_loss22

dqn_network.DeepQNetwork._loss22

private

_lr

dqn_network.DeepQNetwork._lr

private

_memory

dqn_network.DeepQNetwork._memory

private

_memory_counter

dqn_network.DeepQNetwork._memory_counter

private

_memory_size

dqn_network.DeepQNetwork._memory_size

private

_merged

dqn_network.DeepQNetwork._merged

private

_n_actions

dqn_network.DeepQNetwork._n_actions

private

_n_features

dqn_network.DeepQNetwork._n_features

private

_output_tensorboard

dqn_network.DeepQNetwork._output_tensorboard

private

_q_eval22

dqn_network.DeepQNetwork._q_eval22

private

_q_next22

dqn_network.DeepQNetwork._q_next22

private

_q_target11

dqn_network.DeepQNetwork._q_target11

private

_q_target12

dqn_network.DeepQNetwork._q_target12

private

_q_target21

dqn_network.DeepQNetwork._q_target21

private

_q_target22

dqn_network.DeepQNetwork._q_target22

private

_replace_target_iter

dqn_network.DeepQNetwork._replace_target_iter

private

_replace_target_op

dqn_network.DeepQNetwork._replace_target_op

private

_reward

dqn_network.DeepQNetwork._reward

private

_reward_hit

dqn_network.DeepQNetwork._reward_hit

private

_s

dqn_network.DeepQNetwork._s

private

_s_

dqn_network.DeepQNetwork._s_

private

_sess

dqn_network.DeepQNetwork._sess

private

_summary_writer

dqn_network.DeepQNetwork._summary_writer

private

_train_op11

dqn_network.DeepQNetwork._train_op11

private

_train_op12

dqn_network.DeepQNetwork._train_op12

private

_train_op21

dqn_network.DeepQNetwork._train_op21

private

_train_op22

dqn_network.DeepQNetwork._train_op22

private

---

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

• src/ICRA-Firefly-RoboRTS/roborts_decision/dqn_decision/dqn_network.py