Exemplos de ReplayBuffer em Python

Exemplo n.º 1

Arquivo: 4-train-ddpg-cnn-remote.py Projeto: krrish94/challenge-aido1_LF1-baseline-RL-sim-pytorch

env = NormalizeWrapper(env)
env = ImgWrapper(env)  # to make the images from 160x120x3 into 3x160x120
env = ActionWrapper(env)
env = DtRewardWrapper(env)

# Set seeds
seed(args.seed)

state_dim = env.observation_space.shape
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])

# Initialize policy
policy = DDPG(state_dim, action_dim, max_action, net_type="cnn")

replay_buffer = utils.ReplayBuffer(args.replay_buffer_max_size)

# Evaluate untrained policy
evaluations = [evaluate_policy(env, policy)]

exp.metric("rewards", evaluations[0])

total_timesteps = 0
timesteps_since_eval = 0
episode_num = 0
done = True
episode_reward = None
env_counter = 0
while total_timesteps < args.max_timesteps:

    if done:

Exemplo n.º 2

env = gym.make("Duckietown-loop_obstacles-v0")

# Wrappers
env = NormalizeWrapper(env)

# Set seeds
seed(args.seed)

state_dim = env.observation_space.shape
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])

# Initialize policy
policy = DDPG(state_dim, action_dim, max_action, net_type="dense")

replay_buffer = utils.ReplayBuffer()

# Evaluate untrained policy
evaluations = [evaluate_policy(env, policy)]

exp.metric("rewards", evaluations[0])

total_timesteps = 0
timesteps_since_eval = 0
episode_num = 0
done = True
episode_reward = None
env_counter = 0
while total_timesteps < args.max_timesteps:

    if done: