wassname/pytorch-soft-actor-critic
1
0
Fork 0
mirror of https://github.com/wassname/pytorch-soft-actor-critic.git synced 2026-08-07 11:08:21 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update main.py

Browse Source
This commit is contained in:
Pranjal Tandon
2018-12-05 02:08:17 +05:30
committed by GitHub
parent edc0109fad
commit 9d049a4af9
1 changed files with 5 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main.py
+5 -4
View File
@@ -12,8 +12,8 @@ from replay_memory import ReplayMemory
parser = argparse.ArgumentParser(description='PyTorch REINFORCE example')
parser.add_argument('--env-name', default="Pendulum-v0",
help='name of the environment to run')
parser.add_argument('--deterministic', type=bool, default=False,
help='use a deterministic policy (default:False)')
parser.add_argument('--policy', default="Gaussian",
help='algorithm to use: Gaussian | Deterministic')
parser.add_argument('--eval', type=bool, default=False,
help='Evaluate a policy (default:False)')
parser.add_argument('--reparam', type=bool, default=True,
@@ -24,8 +24,8 @@ parser.add_argument('--tau', type=float, default=0.005, metavar='G',
help='target smoothing coefficient(τ) (default: 0.005)')
parser.add_argument('--lr', type=float, default=0.0003, metavar='G',
help='learning rate (default: 0.0003)')
parser.add_argument('--scale_R', type=int, default=5, metavar='G',
help='reward scaling (default: 5)')
parser.add_argument('--alpha', type=float, default=0.2, metavar='G',
help='Entropy regularization coefficient is equivalent to inverse of reward scale. It is much more stable for training. (default: 0.2)')
parser.add_argument('--seed', type=int, default=543, metavar='N',
help='random seed (default: 543)')
parser.add_argument('--batch_size', type=int, default=256, metavar='N',
@@ -78,6 +78,7 @@ for i_episode in itertools.count():
value_loss, critic_1_loss, critic_2_loss, policy_loss = agent.update_parameters(state_batch, action_batch,
reward_batch, next_state_batch,
mask_batch, updates)
writer.add_scalar('loss/value', value_loss, updates)
writer.add_scalar('loss/critic_1', critic_1_loss, updates)
writer.add_scalar('loss/critic_2', critic_2_loss, updates)