Skip to content
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

Add networksettings to reward providers #4982

Merged
merged 9 commits into from
Feb 22, 2021
Merged

Add networksettings to reward providers #4982

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
b68fbdc
add network settings to reward providers
andrewcoh Feb 19, 2021
4f83bbf
update pyramids rnd config
andrewcoh Feb 19, 2021
00eb2b2
remove print statement
andrewcoh Feb 19, 2021
d971b14
fix vail tests
andrewcoh Feb 19, 2021
55ebc79
add comment to RewardSignal structure
andrewcoh Feb 20, 2021
fbb3ebe
set default encoding size to optional int
andrewcoh Feb 20, 2021
cc72294
update documentations in training config
andrewcoh Feb 20, 2021
22db076
update changelog
andrewcoh Feb 20, 2021
b43e20c
raise warning if memory specified in reward providers
andrewcoh Feb 22, 2021
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: 5 additions & 1 deletion config/imitation/CrawlerStatic.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,11 @@ behaviors:
gail:
gamma: 0.99
strength: 1.0
encoding_size: 128
network_settings:
normalize: true
hidden_units: 128
num_layers: 2
vis_encode_type: simple
learning_rate: 0.0003
use_actions: false
use_vail: false
Expand Down
6 changes: 5 additions & 1 deletion config/imitation/FoodCollector.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,11 @@ behaviors:
gail:
gamma: 0.99
strength: 0.1
encoding_size: 128
network_settings:
normalize: false
hidden_units: 128
num_layers: 2
vis_encode_type: simple
learning_rate: 0.0003
use_actions: false
use_vail: false
Expand Down
1 change: 0 additions & 1 deletion config/imitation/Hallway.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ behaviors:
gail:
gamma: 0.99
strength: 0.01
encoding_size: 128
andrewcoh marked this conversation as resolved.
Show resolved Hide resolved
learning_rate: 0.0003
use_actions: false
use_vail: false
Expand Down
6 changes: 5 additions & 1 deletion config/imitation/PushBlock.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,11 @@ behaviors:
gail:
gamma: 0.99
strength: 0.01
encoding_size: 128
network_settings:
normalize: false
hidden_units: 128
num_layers: 2
vis_encode_type: simple
learning_rate: 0.0003
use_actions: false
use_vail: false
Expand Down
4 changes: 2 additions & 2 deletions config/imitation/Pyramids.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,11 +22,11 @@ behaviors:
curiosity:
strength: 0.02
gamma: 0.99
encoding_size: 256
network_settings:
hidden_units: 256
gail:
strength: 0.01
gamma: 0.99
encoding_size: 128
demo_path: Project/Assets/ML-Agents/Examples/Pyramids/Demos/ExpertPyramid.demo
behavioral_cloning:
demo_path: Project/Assets/ML-Agents/Examples/Pyramids/Demos/ExpertPyramid.demo
Expand Down
3 changes: 2 additions & 1 deletion config/ppo/Pyramids.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,8 @@ behaviors:
curiosity:
gamma: 0.99
strength: 0.02
encoding_size: 256
network_settings:
hidden_units: 256
learning_rate: 0.0003
keep_checkpoints: 5
max_steps: 10000000
Expand Down
2 changes: 2 additions & 0 deletions config/ppo/PyramidsRND.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,8 @@ behaviors:
gamma: 0.99
strength: 0.01
encoding_size: 64
network_settings:
hidden_units: 64
learning_rate: 0.0001
keep_checkpoints: 5
max_steps: 3000000
Expand Down
3 changes: 2 additions & 1 deletion config/ppo/VisualPyramids.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,8 @@ behaviors:
curiosity:
gamma: 0.99
strength: 0.01
encoding_size: 256
network_settings:
hidden_units: 256
learning_rate: 0.0003
keep_checkpoints: 5
max_steps: 10000000
Expand Down
1 change: 0 additions & 1 deletion config/sac/Pyramids.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,6 @@ behaviors:
gail:
gamma: 0.99
strength: 0.01
encoding_size: 128
learning_rate: 0.0003
use_actions: true
use_vail: false
Expand Down
1 change: 0 additions & 1 deletion config/sac/VisualPyramids.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,6 @@ behaviors:
gail:
gamma: 0.99
strength: 0.02
encoding_size: 128
learning_rate: 0.0003
use_actions: true
use_vail: false
Expand Down
3 changes: 0 additions & 3 deletions ml-agents/mlagents/trainers/optimizer/torch_optimizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,9 +44,6 @@ def create_reward_signals(self, reward_signal_configs):
:param reward_signal_configs: Reward signal config.
"""
for reward_signal, settings in reward_signal_configs.items():
# Get normalization from policy. Will be replaced by RewardSettings own
# NetworkSettings
settings.normalize = self.policy.normalize
# Name reward signals by string in case we have duplicates later
self.reward_signals[reward_signal.value] = create_reward_provider(
reward_signal, self.policy.behavior_spec, settings
Expand Down
17 changes: 13 additions & 4 deletions ml-agents/mlagents/trainers/settings.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -183,7 +183,7 @@ def to_settings(self) -> type:
class RewardSignalSettings:
gamma: float = 0.99
strength: float = 1.0
normalize: bool = False
network_settings: NetworkSettings = attr.ib(factory=NetworkSettings)
Copy link
Contributor

Choose a reason for hiding this comment

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

I would make this one optional and if it is None, then use the Policy's network settings rather than our own defaults. How does that sound ?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I think I'd prefer to use our defaults since it's possible the policy has significantly more capacity than is needed i.e. the Crawler policy of 3/512 vs what we use for the discriminator 2/128. That being said, I also realize this enables users to specify memory which we probably want to explicitly prevent in the reward providers. cc @ervteng

Copy link
Contributor

Choose a reason for hiding this comment

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

Not opposed to either route, they have their own pros/cons. Either way as long as it's documented it should be fine.
Is getting the Policy settings super ugly?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Im not sure how future proof it is for multi-agent scenarios. We could have different policies to select from. Additionally, we currently create reward signals in the optimizer/torch_optimizer.py and in the future i think it will be necessary to remove the policy from the optimizer (also for multiagent) in which case this would need to be addressed by either keeping the policy around/moving the creation of the reward provider. My vote is for default network settings


@staticmethod
def structure(d: Mapping, t: type) -> Any:
Expand All @@ -199,28 +199,37 @@ def structure(d: Mapping, t: type) -> Any:
enum_key = RewardSignalType(key)
t = enum_key.to_settings()
d_final[enum_key] = strict_to_cls(val, t)
if "encoding_size" in val:
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Backward compatible with old configs

Copy link
Contributor

Choose a reason for hiding this comment

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

Can you ad a comment around this code so we will remember it ?

logger.warning(
"'encoding_size' was deprecated for RewardSignals. Please use network_settings."
)
# If network settings was not specified, use the encoding size. Otherwise, use hidden_units
if "network_settings" not in val:
d_final[enum_key].network_settings.hidden_units = val[
"encoding_size"
]
return d_final


@attr.s(auto_attribs=True)
class GAILSettings(RewardSignalSettings):
encoding_size: int = 64
learning_rate: float = 3e-4
encoding_size: int = 0
andrewcoh marked this conversation as resolved.
Show resolved Hide resolved
use_actions: bool = False
use_vail: bool = False
demo_path: str = attr.ib(kw_only=True)


@attr.s(auto_attribs=True)
class CuriositySettings(RewardSignalSettings):
encoding_size: int = 64
learning_rate: float = 3e-4
encoding_size: int = 0


@attr.s(auto_attribs=True)
class RNDSettings(RewardSignalSettings):
encoding_size: int = 64
learning_rate: float = 1e-4
encoding_size: int = 0


# SAMPLERS #############################################################################
Expand Down
19 changes: 8 additions & 11 deletions ml-agents/mlagents/trainers/torch/components/reward_providers/curiosity_reward_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,6 @@
from mlagents.trainers.torch.utils import ModelUtils
from mlagents.trainers.torch.networks import NetworkBody
from mlagents.trainers.torch.layers import LinearEncoder, linear_layer
from mlagents.trainers.settings import NetworkSettings, EncoderType
from mlagents.trainers.trajectory import ObsUtil


Expand Down Expand Up @@ -70,21 +69,16 @@ class CuriosityNetwork(torch.nn.Module):
def __init__(self, specs: BehaviorSpec, settings: CuriositySettings) -> None:
super().__init__()
self._action_spec = specs.action_spec
state_encoder_settings = NetworkSettings(
normalize=False,
hidden_units=settings.encoding_size,
num_layers=2,
vis_encode_type=EncoderType.SIMPLE,
memory=None,
)

state_encoder_settings = settings.network_settings
self._state_encoder = NetworkBody(
specs.observation_specs, state_encoder_settings
)

self._action_flattener = ActionFlattener(self._action_spec)

self.inverse_model_action_encoding = torch.nn.Sequential(
LinearEncoder(2 * settings.encoding_size, 1, 256)
LinearEncoder(2 * state_encoder_settings.hidden_units, 1, 256)
)

if self._action_spec.continuous_size > 0:
Expand All @@ -98,9 +92,12 @@ def __init__(self, specs: BehaviorSpec, settings: CuriositySettings) -> None:

self.forward_model_next_state_prediction = torch.nn.Sequential(
LinearEncoder(
settings.encoding_size + self._action_flattener.flattened_size, 1, 256
state_encoder_settings.hidden_units
+ self._action_flattener.flattened_size,
1,
256,
),
linear_layer(256, settings.encoding_size),
linear_layer(256, state_encoder_settings.hidden_units),
)

def get_current_state(self, mini_batch: AgentBuffer) -> torch.Tensor:
Expand Down
12 changes: 3 additions & 9 deletions ml-agents/mlagents/trainers/torch/components/reward_providers/gail_reward_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@
from mlagents.trainers.torch.action_flattener import ActionFlattener
from mlagents.trainers.torch.networks import NetworkBody
from mlagents.trainers.torch.layers import linear_layer, Initialization
from mlagents.trainers.settings import NetworkSettings, EncoderType
from mlagents.trainers.demo_loader import demo_to_buffer
from mlagents.trainers.trajectory import ObsUtil

Expand Down Expand Up @@ -75,13 +74,8 @@ def __init__(self, specs: BehaviorSpec, settings: GAILSettings) -> None:
self._use_vail = settings.use_vail
self._settings = settings

encoder_settings = NetworkSettings(
normalize=settings.normalize,
hidden_units=settings.encoding_size,
num_layers=2,
vis_encode_type=EncoderType.SIMPLE,
memory=None,
)
encoder_settings = settings.network_settings
print(encoder_settings.hidden_units)
andrewcoh marked this conversation as resolved.
Show resolved Hide resolved
self._action_flattener = ActionFlattener(specs.action_spec)
unencoded_size = (
self._action_flattener.flattened_size + 1 if settings.use_actions else 0
Expand All @@ -90,7 +84,7 @@ def __init__(self, specs: BehaviorSpec, settings: GAILSettings) -> None:
specs.observation_specs, encoder_settings, unencoded_size
)

estimator_input_size = settings.encoding_size
estimator_input_size = encoder_settings.hidden_units
if settings.use_vail:
estimator_input_size = self.z_size
self._z_sigma = torch.nn.Parameter(
Expand Down
9 changes: 1 addition & 8 deletions ml-agents/mlagents/trainers/torch/components/reward_providers/rnd_reward_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,6 @@
from mlagents_envs.base_env import BehaviorSpec
from mlagents.trainers.torch.utils import ModelUtils
from mlagents.trainers.torch.networks import NetworkBody
from mlagents.trainers.settings import NetworkSettings, EncoderType
from mlagents.trainers.trajectory import ObsUtil


Expand Down Expand Up @@ -58,13 +57,7 @@ class RNDNetwork(torch.nn.Module):

def __init__(self, specs: BehaviorSpec, settings: RNDSettings) -> None:
super().__init__()
state_encoder_settings = NetworkSettings(
normalize=True,
hidden_units=settings.encoding_size,
num_layers=3,
vis_encode_type=EncoderType.SIMPLE,
memory=None,
)
state_encoder_settings = settings.network_settings
self._encoder = NetworkBody(specs.observation_specs, state_encoder_settings)

def forward(self, mini_batch: AgentBuffer) -> torch.Tensor:
Expand Down