# Every save_ckpt_interval, Check if there is any checkpoint.
# If there is, load checkpoint and continue training
# Need to specify the i_episode of the checkpoint intended to load
# if i_epoch % save_ckpt_interval == 0 and os.path.isfile(os.path.join(ckpt_dir, "ckpt_eps%d.pt" % i_epoch)):
# policy_net, value_net_in, value_net_ex, valuenet_in_optimizer, valuenet_ex_optimizer,\
# simhash, training_info = \
# load_checkpoint(ckpt_dir, i_epoch, layer_sizes, input_size, device=device)
# print("\n\tCheckpoint successfully loaded!\n")
# To record episode stats
episode_durations = []
episode_rewards = []
# Use value net in evaluation mode when collecting trajectories
value_net_in.eval()
value_net_ex.eval()
###################################################################
# Collect trajectories
print("\n\n\tCollecting %d episodes: " % (batch_size))
for i_episode in tqdm(range(batch_size)): # Use tqdm to show progress bar
# Keep track of the running reward
running_reward = 0
# Initialize the environment and state
current_state = env.reset()
finished_rendering_this_epoch = False
# Every save_ckpt_interval, Check if there is any checkpoint.
# If there is, load checkpoint and continue training
# Need to specify the i_episode of the checkpoint intended to load
if i_epoch % save_ckpt_interval == 0 and os.path.isfile(
os.path.join(ckpt_dir, "ckpt_eps%d.pt" % i_epoch)):
policy_net, value_net, valuenet_optimizer, training_info = \
load_checkpoint(ckpt_dir, i_epoch, layer_sizes, input_size, device=device)
# To record episode stats
episode_durations = []
episode_rewards = []
# Use value net in evaluation mode when collecting trajectories
value_net.eval()
###################################################################
# Collect trajectories
print("\n\n\tCollecting %d episodes: " % (batch_size))
for i_episode in tqdm(range(batch_size)): # Use tqdm to show progress bar
# Keep track of the running reward
running_reward = 0
# Initialize the environment and state
current_state = env.reset()
# Estimate the value of the initial state