def main(game, _seed, _run):
torch.manual_seed(_seed)
game = lower_under_to_upper(game) + 'NoFrameskip-v4'
env = gym.make(game)
env = wrap_deepmind(env)
input_space = env.observation_space
num_actions = env.action_space.n
agent_params = DeepQAgentParams()
add_params(params=agent_params, prefix='agent')
add_params(params=agent_params.optimizer_params, prefix='opt')
add_epsilon_params(params=agent_params)
agent_params.obs_filter = AtariObservationFilter()
input_space = agent_params.obs_filter.output_space(input_space)
agent_params.sacred_run = _run
agent_params.env = env
agent_params.mode = 'train'
online_q_net = build_net(input_shape=input_space.shape,
num_actions=num_actions)
target_q_net = build_net(input_shape=input_space.shape,
num_actions=num_actions)
agent_params.online_q_net = online_q_net
agent_params.target_q_net = target_q_net
agent = agent_params.make_agent()
agent.run()
def main(_seed, _run):
torch.manual_seed(_seed)
env = build_env()
input_shape = env.observation_space.shape
num_actions = env.action_space.n
agent_params = DeepQAgentParams()
add_params(params=agent_params, prefix='agent')
add_params(params=agent_params.optimizer_params, prefix='opt')
add_epsilon_params(params=agent_params)
agent_params.sacred_run = _run
agent_params.env = env
agent_params.mode = 'train'
online_q_net = build_net(input_shape=input_shape, num_actions=num_actions)
target_q_net = build_net(input_shape=input_shape, num_actions=num_actions)
agent_params.online_q_net = online_q_net
agent_params.target_q_net = target_q_net
agent = agent_params.make_agent()
agent.run()