Skip to content

[CI] Better hyperparameters for Pyramids-SAC, WalkerStatic-SAC, and Reacher-PPO #4154

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 12 commits into from
Jun 29, 2020
Merged

[CI] Better hyperparameters for Pyramids-SAC, WalkerStatic-SAC, and Reacher-PPO #4154

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
585d884
Better hyperparameters for WalkerStatic
Jun 16, 2020
b0fa7f2
Lower beta for Reacher
Jun 16, 2020
34b88ce
Experimental Pyramids config
Jun 17, 2020
2e4f45a
Bigger networks
Jun 17, 2020
6106b09
Use curiosity instead of GAIL
Jun 17, 2020
9a1d995
Bigger buffer size
Jun 17, 2020
e24d49d
Switch back to GAIL
Jun 17, 2020
ce74afd
Reduce Reacher batch size
Jun 18, 2020
583489e
Final Pyramids config
Jun 18, 2020
2d94512
Reduce Pyramids buffer size
Jun 24, 2020
c568601
Increase buffer size for testing only
Jun 24, 2020
cf9af0d
Shorten Pyramids run
Jun 25, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions config/ppo/Reacher.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,10 +2,10 @@ behaviors:
Reacher:
trainer_type: ppo
hyperparameters:
batch_size: 2024
buffer_size: 20240
batch_size: 512
buffer_size: 20480
learning_rate: 0.0003
beta: 0.005
beta: 0.001
epsilon: 0.2
lambd: 0.95
num_epoch: 3
Expand Down
14 changes: 7 additions & 7 deletions config/sac/Pyramids.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,32 +5,32 @@ behaviors:
learning_rate: 0.0003
learning_rate_schedule: constant
batch_size: 128
buffer_size: 500000
buffer_init_steps: 10000
buffer_size: 2000000
buffer_init_steps: 1000
tau: 0.01
steps_per_update: 10.0
save_replay_buffer: false
init_entcoef: 0.01
reward_signal_steps_per_update: 10.0
network_settings:
normalize: false
hidden_units: 256
num_layers: 2
hidden_units: 512
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did you run with just the larger buffer?

Copy link
Contributor Author

@ervteng ervteng Jun 23, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, it is more stable but it peaks out much earlier (around 1.2-ish) and the losses continue to climb.

num_layers: 3
vis_encode_type: simple
reward_signals:
extrinsic:
gamma: 0.99
gamma: 0.995
strength: 2.0
gail:
gamma: 0.99
strength: 0.02
strength: 0.01
encoding_size: 128
learning_rate: 0.0003
use_actions: true
use_vail: false
demo_path: Project/Assets/ML-Agents/Examples/Pyramids/Demos/ExpertPyramid.demo
keep_checkpoints: 5
max_steps: 10000000
max_steps: 3000000
time_horizon: 128
summary_freq: 30000
threaded: true
10 changes: 5 additions & 5 deletions config/sac/WalkerStatic.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,8 +4,8 @@ behaviors:
hyperparameters:
learning_rate: 0.0003
learning_rate_schedule: constant
batch_size: 256
buffer_size: 500000
batch_size: 1024
buffer_size: 2000000
buffer_init_steps: 0
tau: 0.005
steps_per_update: 30.0
Expand All @@ -14,15 +14,15 @@ behaviors:
reward_signal_steps_per_update: 30.0
network_settings:
normalize: true
hidden_units: 512
num_layers: 4
hidden_units: 256
num_layers: 3
vis_encode_type: simple
reward_signals:
extrinsic:
gamma: 0.995
strength: 1.0
keep_checkpoints: 5
max_steps: 20000000
max_steps: 15000000
time_horizon: 1000
summary_freq: 30000
threaded: true