def train_double_dqn(env, num_actions):
results_dir = './results/double_dqn/' + game
training_epsilon = 0.01
test_epsilon = 0.001
frame_history = 1
dqn = atari_dqn.AtariDQN(frame_history, num_actions)
agent = dq_learner_pc.DQLearner(dqn,
num_actions,
frame_history=frame_history,
epsilon_end=training_epsilon)
train(agent, env, test_epsilon, results_dir)
def train_dqn(env, num_actions):
results_dir = './results/dqn/' + game
training_epsilon = 0.1
test_epsilon = 0.05
frame_history = 1
dqn = atari_dqn.AtariDQN(frame_history, num_actions, shared_bias=False)
agent = dq_learner_pc.DQLearner(dqn,
num_actions,
target_copy_freq=10000,
epsilon_end=training_epsilon,
double=False,
frame_history=frame_history)
train(agent, env, test_epsilon, results_dir)
def train_double_dqn(env, num_actions):
results_dir = './results/double_dqn/' + game
training_epsilon = 0.01
test_epsilon = 0.001
frame_history = 4
cts_size = (42, 42, 8)
enc_func = atari_encoder.encode_state
dqn = atari_dqn.AtariDQN(frame_history, num_actions)
agent = dq_learner_pc.DQLearner(dqn,
num_actions,
frame_history=frame_history,
epsilon_end=training_epsilon,
state_encoder=enc_func,
cts_size=cts_size)
train(agent, env, test_epsilon, results_dir, max_episode_steps=None)
def train_dqn(env, num_actions):
results_dir = './results/dqn/' + game
training_epsilon = 0.1
test_epsilon = 0.05
frame_history = 4
cts_size = (42, 42, 8)
enc_func = atari_encoder.encode_state
dqn = atari_dqn.AtariDQN(frame_history, num_actions, shared_bias=False)
agent = dq_learner_pc.DQLearner(dqn,
num_actions,
target_copy_freq=10000,
epsilon_end=training_epsilon,
double=False,
frame_history=frame_history,
state_encoder=enc_func,
cts_size=cts_size)
train(agent, env, test_epsilon, results_dir, max_episode_steps=None)