Prioritized Experienced Replay DQN

This is the implementation of the Prioritized Experienced Replay based Deep Q Learning.

Classes and Functions

class per_dqn.Agent(state_size, action_size, per=True)

Bases: object

act(state)

compute_TDerror(transition)

load(name)

remember(state, action, reward, next_state, done)

replay(batch_size)

save(name)

target_train()

class per_dqn.PER(max_size)

Bases: object

Prioritized replay memory using binary heap

add(transition, TDerror)

batch(n)

is_full()

size()

class per_dqn.Replay_Memory(max_size)

Bases: object

Standard replay memory sampled uniformly

add(transition)

batch(n)

is_full()

size()