dopamine icon indicating copy to clipboard operation
dopamine copied to clipboard

Reproduct breakout score reported in baselines folder

Open tsachiblau opened this issue 5 years ago • 0 comments

Hello,

I'm trying to reproduct breakout score reported in baselines folder. I saw that the score reach around 100. I am running v0 and reaching around 40 at best.

this is my GIN file

HierarchyDQNAgent.gamma = 0.99 HierarchyDQNAgent.update_horizon = 1 HierarchyDQNAgent.min_replay_history = 200000 # agent steps HierarchyDQNAgent.update_period = 4 HierarchyDQNAgent.target_update_period = 40000 # agent steps HierarchyDQNAgent.epsilon_train = 0.1 HierarchyDQNAgent.epsilon_eval = 0.01 HierarchyDQNAgent.epsilon_decay_period = 4000000 # agent steps HierarchyDQNAgent.tf_device = '/gpu:0' # use '/cpu:*' for non-GPU version HierarchyDQNAgent.optimizer = @tf.train.RMSPropOptimizer()

tf.train.RMSPropOptimizer.learning_rate = 0.00025 tf.train.RMSPropOptimizer.decay = 0.95 tf.train.RMSPropOptimizer.momentum = 0.0 tf.train.RMSPropOptimizer.epsilon = 0.00001 tf.train.RMSPropOptimizer.centered = True

atari_lib.create_atari_environment.game_name = 'Breakout' atari_lib.create_atari_environment.sticky_actions = True create_agent.agent_name = 'hierarchy' Runner.num_iterations = 300 Runner.training_steps = 250000 # agent steps Runner.evaluation_steps = 125000 # agent steps Runner.max_steps_per_episode = 27000 # agent steps

AtariPreprocessing.terminal_on_life_loss = True

WrappedReplayBuffer.replay_capacity = 1000000 WrappedReplayBuffer.batch_size = 32

any idea what I should change?

Thanks, Tsachi

tsachiblau avatar May 25 '19 15:05 tsachiblau