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

fix: remove eval() python method #265

Merged
merged 2 commits into from
Apr 15, 2024
Merged

fix: remove eval() python method #265

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ef09779
fix: remove eval() python method
michele-milesi Apr 14, 2024
7c18bab
fix: dreamer_v3 doapp configs
michele-milesi Apr 14, 2024
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
2 changes: 1 addition & 1 deletion howto/learn_in_diambra.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ env:
diambra_settings:
characters: Kasumi
step_ratio: 5
role: diambra.arena.Roles.P1
role: P1
diambra_wrappers:
normalize_reward: True
normalization_factor: 0.3
Expand Down
7 changes: 4 additions & 3 deletions howto/register_external_algorithm.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,6 +96,7 @@ from math import prod
from typing import Any, Dict, List, Optional, Sequence, Tuple

import gymnasium
import hydra
import torch
import torch.nn as nn
from lightning import Fabric
Expand Down Expand Up @@ -198,7 +199,7 @@ class PPOAgent(nn.Module):
mlp_keys,
encoder_cfg.dense_units,
encoder_cfg.mlp_layers,
eval(encoder_cfg.dense_act),
hydra.utils.get_class(encoder_cfg.dense_act),
encoder_cfg.layer_norm,
)
if mlp_keys is not None and len(mlp_keys) > 0
Expand All @@ -210,7 +211,7 @@ class PPOAgent(nn.Module):
input_dims=features_dim,
output_dim=1,
hidden_sizes=[critic_cfg.dense_units] * critic_cfg.mlp_layers,
activation=eval(critic_cfg.dense_act),
activation=hydra.utils.get_class(critic_cfg.dense_act),
norm_layer=[nn.LayerNorm for _ in range(critic_cfg.mlp_layers)] if critic_cfg.layer_norm else None,
norm_args=(
[{"normalized_shape": critic_cfg.dense_units} for _ in range(critic_cfg.mlp_layers)]
Expand All @@ -223,7 +224,7 @@ class PPOAgent(nn.Module):
input_dims=features_dim,
output_dim=None,
hidden_sizes=[actor_cfg.dense_units] * actor_cfg.mlp_layers,
activation=eval(actor_cfg.dense_act),
activation=hydra.utils.get_class(actor_cfg.dense_act),
flatten_dim=None,
norm_layer=[nn.LayerNorm] * actor_cfg.mlp_layers if actor_cfg.layer_norm else None,
norm_args=(
Expand Down
7 changes: 4 additions & 3 deletions howto/register_new_algorithm.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,6 +95,7 @@ from math import prod
from typing import Any, Dict, List, Optional, Sequence, Tuple

import gymnasium
import hydra
import torch
import torch.nn as nn
from lightning import Fabric
Expand Down Expand Up @@ -197,7 +198,7 @@ class PPOAgent(nn.Module):
mlp_keys,
encoder_cfg.dense_units,
encoder_cfg.mlp_layers,
eval(encoder_cfg.dense_act),
hydra.utils.get_class(encoder_cfg.dense_act),
encoder_cfg.layer_norm,
)
if mlp_keys is not None and len(mlp_keys) > 0
Expand All @@ -209,7 +210,7 @@ class PPOAgent(nn.Module):
input_dims=features_dim,
output_dim=1,
hidden_sizes=[critic_cfg.dense_units] * critic_cfg.mlp_layers,
activation=eval(critic_cfg.dense_act),
activation=hydra.utils.get_class(critic_cfg.dense_act),
norm_layer=[nn.LayerNorm for _ in range(critic_cfg.mlp_layers)] if critic_cfg.layer_norm else None,
norm_args=(
[{"normalized_shape": critic_cfg.dense_units} for _ in range(critic_cfg.mlp_layers)]
Expand All @@ -222,7 +223,7 @@ class PPOAgent(nn.Module):
input_dims=features_dim,
output_dim=None,
hidden_sizes=[actor_cfg.dense_units] * actor_cfg.mlp_layers,
activation=eval(actor_cfg.dense_act),
activation=hydra.utils.get_class(actor_cfg.dense_act),
flatten_dim=None,
norm_layer=[nn.LayerNorm] * actor_cfg.mlp_layers if actor_cfg.layer_norm else None,
norm_args=(
Expand Down
7 changes: 4 additions & 3 deletions sheeprl/algos/a2c/agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@
from typing import Any, Dict, List, Optional, Sequence, Tuple

import gymnasium
import hydra
import torch
import torch.nn as nn
from lightning import Fabric
Expand Down Expand Up @@ -72,7 +73,7 @@ def __init__(
mlp_keys,
encoder_cfg.dense_units,
encoder_cfg.mlp_layers,
eval(encoder_cfg.dense_act),
hydra.utils.get_class(encoder_cfg.dense_act),
encoder_cfg.layer_norm,
)
if mlp_keys is not None and len(mlp_keys) > 0
Expand All @@ -85,7 +86,7 @@ def __init__(
input_dims=features_dim,
output_dim=1,
hidden_sizes=[critic_cfg.dense_units] * critic_cfg.mlp_layers,
activation=eval(critic_cfg.dense_act),
activation=hydra.utils.get_class(critic_cfg.dense_act),
norm_layer=[nn.LayerNorm for _ in range(critic_cfg.mlp_layers)] if critic_cfg.layer_norm else None,
norm_args=(
[{"normalized_shape": critic_cfg.dense_units} for _ in range(critic_cfg.mlp_layers)]
Expand All @@ -99,7 +100,7 @@ def __init__(
input_dims=features_dim,
output_dim=None,
hidden_sizes=[actor_cfg.dense_units] * actor_cfg.mlp_layers,
activation=eval(actor_cfg.dense_act),
activation=hydra.utils.get_class(actor_cfg.dense_act),
flatten_dim=None,
norm_layer=[nn.LayerNorm] * actor_cfg.mlp_layers if actor_cfg.layer_norm else None,
norm_args=(
Expand Down
22 changes: 11 additions & 11 deletions sheeprl/algos/dreamer_v1/agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -373,7 +373,7 @@ def build_agent(
image_size=obs_space[cfg.algo.cnn_keys.encoder[0]].shape[-2:],
channels_multiplier=world_model_cfg.encoder.cnn_channels_multiplier,
layer_norm=False,
activation=eval(world_model_cfg.encoder.cnn_act),
activation=hydra.utils.get_class(world_model_cfg.encoder.cnn_act),
)
if cfg.algo.cnn_keys.encoder is not None and len(cfg.algo.cnn_keys.encoder) > 0
else None
Expand All @@ -384,7 +384,7 @@ def build_agent(
input_dims=[obs_space[k].shape[0] for k in cfg.algo.mlp_keys.encoder],
mlp_layers=world_model_cfg.encoder.mlp_layers,
dense_units=world_model_cfg.encoder.dense_units,
activation=eval(world_model_cfg.encoder.dense_act),
activation=hydra.utils.get_class(world_model_cfg.encoder.dense_act),
layer_norm=False,
)
if cfg.algo.mlp_keys.encoder is not None and len(cfg.algo.mlp_keys.encoder) > 0
Expand All @@ -394,22 +394,22 @@ def build_agent(
recurrent_model = RecurrentModel(
input_size=sum(actions_dim) + world_model_cfg.stochastic_size,
recurrent_state_size=world_model_cfg.recurrent_model.recurrent_state_size,
activation=eval(world_model_cfg.recurrent_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.recurrent_model.dense_act),
)
representation_model = MLP(
input_dims=(
world_model_cfg.recurrent_model.recurrent_state_size + encoder.cnn_output_dim + encoder.mlp_output_dim
),
output_dim=world_model_cfg.stochastic_size * 2,
hidden_sizes=[world_model_cfg.representation_model.hidden_size],
activation=eval(world_model_cfg.representation_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.representation_model.dense_act),
flatten_dim=None,
)
transition_model = MLP(
input_dims=world_model_cfg.recurrent_model.recurrent_state_size,
output_dim=world_model_cfg.stochastic_size * 2,
hidden_sizes=[world_model_cfg.transition_model.hidden_size],
activation=eval(world_model_cfg.transition_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.transition_model.dense_act),
flatten_dim=None,
)
rssm = RSSM(
Expand All @@ -427,7 +427,7 @@ def build_agent(
latent_state_size=latent_state_size,
cnn_encoder_output_dim=cnn_encoder.output_dim,
image_size=obs_space[cfg.algo.cnn_keys.decoder[0]].shape[-2:],
activation=eval(world_model_cfg.observation_model.cnn_act),
activation=hydra.utils.get_class(world_model_cfg.observation_model.cnn_act),
layer_norm=False,
)
if cfg.algo.cnn_keys.decoder is not None and len(cfg.algo.cnn_keys.decoder) > 0
Expand All @@ -440,7 +440,7 @@ def build_agent(
latent_state_size=latent_state_size,
mlp_layers=world_model_cfg.observation_model.mlp_layers,
dense_units=world_model_cfg.observation_model.dense_units,
activation=eval(world_model_cfg.observation_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.observation_model.dense_act),
layer_norm=False,
)
if cfg.algo.mlp_keys.decoder is not None and len(cfg.algo.mlp_keys.decoder) > 0
Expand All @@ -451,15 +451,15 @@ def build_agent(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[world_model_cfg.reward_model.dense_units] * world_model_cfg.reward_model.mlp_layers,
activation=eval(world_model_cfg.reward_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.reward_model.dense_act),
flatten_dim=None,
)
if world_model_cfg.use_continues:
continue_model = MLP(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[world_model_cfg.discount_model.dense_units] * world_model_cfg.discount_model.mlp_layers,
activation=eval(world_model_cfg.discount_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.discount_model.dense_act),
flatten_dim=None,
)
world_model = WorldModel(
Expand All @@ -478,7 +478,7 @@ def build_agent(
min_std=actor_cfg.min_std,
mlp_layers=actor_cfg.mlp_layers,
dense_units=actor_cfg.dense_units,
activation=eval(actor_cfg.dense_act),
activation=hydra.utils.get_class(actor_cfg.dense_act),
distribution_cfg=cfg.distribution,
layer_norm=False,
expl_amount=actor_cfg.expl_amount,
Expand All @@ -489,7 +489,7 @@ def build_agent(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[critic_cfg.dense_units] * critic_cfg.mlp_layers,
activation=eval(critic_cfg.dense_act),
activation=hydra.utils.get_class(critic_cfg.dense_act),
flatten_dim=None,
)
actor.apply(init_weights)
Expand Down
20 changes: 10 additions & 10 deletions sheeprl/algos/dreamer_v2/agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -883,7 +883,7 @@ def build_agent(
image_size=obs_space[cfg.algo.cnn_keys.encoder[0]].shape[-2:],
channels_multiplier=world_model_cfg.encoder.cnn_channels_multiplier,
layer_norm=world_model_cfg.encoder.layer_norm,
activation=eval(world_model_cfg.encoder.cnn_act),
activation=hydra.utils.get_class(world_model_cfg.encoder.cnn_act),
)
if cfg.algo.cnn_keys.encoder is not None and len(cfg.algo.cnn_keys.encoder) > 0
else None
Expand All @@ -894,7 +894,7 @@ def build_agent(
input_dims=[obs_space[k].shape[0] for k in cfg.algo.mlp_keys.encoder],
mlp_layers=world_model_cfg.encoder.mlp_layers,
dense_units=world_model_cfg.encoder.dense_units,
activation=eval(world_model_cfg.encoder.dense_act),
activation=hydra.utils.get_class(world_model_cfg.encoder.dense_act),
layer_norm=world_model_cfg.encoder.layer_norm,
)
if cfg.algo.mlp_keys.encoder is not None and len(cfg.algo.mlp_keys.encoder) > 0
Expand All @@ -911,7 +911,7 @@ def build_agent(
),
output_dim=stochastic_size,
hidden_sizes=[world_model_cfg.representation_model.hidden_size],
activation=eval(world_model_cfg.representation_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.representation_model.dense_act),
flatten_dim=None,
norm_layer=[nn.LayerNorm] if world_model_cfg.representation_model.layer_norm else None,
norm_args=(
Expand All @@ -924,7 +924,7 @@ def build_agent(
input_dims=world_model_cfg.recurrent_model.recurrent_state_size,
output_dim=stochastic_size,
hidden_sizes=[world_model_cfg.transition_model.hidden_size],
activation=eval(world_model_cfg.transition_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.transition_model.dense_act),
flatten_dim=None,
norm_layer=[nn.LayerNorm] if world_model_cfg.transition_model.layer_norm else None,
norm_args=(
Expand All @@ -948,7 +948,7 @@ def build_agent(
latent_state_size=latent_state_size,
cnn_encoder_output_dim=cnn_encoder.output_dim,
image_size=obs_space[cfg.algo.cnn_keys.decoder[0]].shape[-2:],
activation=eval(world_model_cfg.observation_model.cnn_act),
activation=hydra.utils.get_class(world_model_cfg.observation_model.cnn_act),
layer_norm=world_model_cfg.observation_model.layer_norm,
)
if cfg.algo.cnn_keys.decoder is not None and len(cfg.algo.cnn_keys.decoder) > 0
Expand All @@ -961,7 +961,7 @@ def build_agent(
latent_state_size=latent_state_size,
mlp_layers=world_model_cfg.observation_model.mlp_layers,
dense_units=world_model_cfg.observation_model.dense_units,
activation=eval(world_model_cfg.observation_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.observation_model.dense_act),
layer_norm=world_model_cfg.observation_model.layer_norm,
)
if cfg.algo.mlp_keys.decoder is not None and len(cfg.algo.mlp_keys.decoder) > 0
Expand All @@ -972,7 +972,7 @@ def build_agent(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[world_model_cfg.reward_model.dense_units] * world_model_cfg.reward_model.mlp_layers,
activation=eval(world_model_cfg.reward_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.reward_model.dense_act),
flatten_dim=None,
norm_layer=(
[nn.LayerNorm for _ in range(world_model_cfg.reward_model.mlp_layers)]
Expand All @@ -993,7 +993,7 @@ def build_agent(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[world_model_cfg.discount_model.dense_units] * world_model_cfg.discount_model.mlp_layers,
activation=eval(world_model_cfg.discount_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.discount_model.dense_act),
flatten_dim=None,
norm_layer=(
[nn.LayerNorm for _ in range(world_model_cfg.discount_model.mlp_layers)]
Expand Down Expand Up @@ -1025,15 +1025,15 @@ def build_agent(
min_std=actor_cfg.min_std,
mlp_layers=actor_cfg.mlp_layers,
dense_units=actor_cfg.dense_units,
activation=eval(actor_cfg.dense_act),
activation=hydra.utils.get_class(actor_cfg.dense_act),
distribution_cfg=cfg.distribution,
layer_norm=actor_cfg.layer_norm,
)
critic = MLP(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[critic_cfg.dense_units] * critic_cfg.mlp_layers,
activation=eval(critic_cfg.dense_act),
activation=hydra.utils.get_class(critic_cfg.dense_act),
flatten_dim=None,
norm_layer=[nn.LayerNorm for _ in range(critic_cfg.mlp_layers)] if critic_cfg.layer_norm else None,
norm_args=(
Expand Down
20 changes: 10 additions & 10 deletions sheeprl/algos/dreamer_v3/agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -986,7 +986,7 @@ def build_agent(
channels_multiplier=world_model_cfg.encoder.cnn_channels_multiplier,
layer_norm_cls=hydra.utils.get_class(world_model_cfg.encoder.cnn_layer_norm.cls),
layer_norm_kw=world_model_cfg.encoder.cnn_layer_norm.kw,
activation=eval(world_model_cfg.encoder.cnn_act),
activation=hydra.utils.get_class(world_model_cfg.encoder.cnn_act),
stages=cnn_stages,
)
if cfg.algo.cnn_keys.encoder is not None and len(cfg.algo.cnn_keys.encoder) > 0
Expand All @@ -998,7 +998,7 @@ def build_agent(
input_dims=[obs_space[k].shape[0] for k in cfg.algo.mlp_keys.encoder],
mlp_layers=world_model_cfg.encoder.mlp_layers,
dense_units=world_model_cfg.encoder.dense_units,
activation=eval(world_model_cfg.encoder.dense_act),
activation=hydra.utils.get_class(world_model_cfg.encoder.dense_act),
layer_norm_cls=hydra.utils.get_class(world_model_cfg.encoder.mlp_layer_norm.cls),
layer_norm_kw=world_model_cfg.encoder.mlp_layer_norm.kw,
)
Expand All @@ -1022,7 +1022,7 @@ def build_agent(
input_dims=represention_model_input_size,
output_dim=stochastic_size,
hidden_sizes=[world_model_cfg.representation_model.hidden_size],
activation=eval(world_model_cfg.representation_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.representation_model.dense_act),
layer_args={"bias": representation_ln_cls == nn.Identity},
flatten_dim=None,
norm_layer=[representation_ln_cls],
Expand All @@ -1038,7 +1038,7 @@ def build_agent(
input_dims=recurrent_state_size,
output_dim=stochastic_size,
hidden_sizes=[world_model_cfg.transition_model.hidden_size],
activation=eval(world_model_cfg.transition_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.transition_model.dense_act),
layer_args={"bias": transition_ln_cls == nn.Identity},
flatten_dim=None,
norm_layer=[transition_ln_cls],
Expand Down Expand Up @@ -1072,7 +1072,7 @@ def build_agent(
latent_state_size=latent_state_size,
cnn_encoder_output_dim=cnn_encoder.output_dim,
image_size=obs_space[cfg.algo.cnn_keys.decoder[0]].shape[-2:],
activation=eval(world_model_cfg.observation_model.cnn_act),
activation=hydra.utils.get_class(world_model_cfg.observation_model.cnn_act),
layer_norm_cls=hydra.utils.get_class(world_model_cfg.observation_model.cnn_layer_norm.cls),
layer_norm_kw=world_model_cfg.observation_model.mlp_layer_norm.kw,
stages=cnn_stages,
Expand All @@ -1087,7 +1087,7 @@ def build_agent(
latent_state_size=latent_state_size,
mlp_layers=world_model_cfg.observation_model.mlp_layers,
dense_units=world_model_cfg.observation_model.dense_units,
activation=eval(world_model_cfg.observation_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.observation_model.dense_act),
layer_norm_cls=hydra.utils.get_class(world_model_cfg.observation_model.mlp_layer_norm.cls),
layer_norm_kw=world_model_cfg.observation_model.mlp_layer_norm.kw,
)
Expand All @@ -1101,7 +1101,7 @@ def build_agent(
input_dims=latent_state_size,
output_dim=world_model_cfg.reward_model.bins,
hidden_sizes=[world_model_cfg.reward_model.dense_units] * world_model_cfg.reward_model.mlp_layers,
activation=eval(world_model_cfg.reward_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.reward_model.dense_act),
layer_args={"bias": reward_ln_cls == nn.Identity},
flatten_dim=None,
norm_layer=reward_ln_cls,
Expand All @@ -1116,7 +1116,7 @@ def build_agent(
input_dims=latent_state_size,
output_dim=1,
hidden_sizes=[world_model_cfg.discount_model.dense_units] * world_model_cfg.discount_model.mlp_layers,
activation=eval(world_model_cfg.discount_model.dense_act),
activation=hydra.utils.get_class(world_model_cfg.discount_model.dense_act),
layer_args={"bias": discount_ln_cls == nn.Identity},
flatten_dim=None,
norm_layer=discount_ln_cls,
Expand All @@ -1141,7 +1141,7 @@ def build_agent(
init_std=actor_cfg.init_std,
min_std=actor_cfg.min_std,
dense_units=actor_cfg.dense_units,
activation=eval(actor_cfg.dense_act),
activation=hydra.utils.get_class(actor_cfg.dense_act),
mlp_layers=actor_cfg.mlp_layers,
distribution_cfg=cfg.distribution,
layer_norm_cls=hydra.utils.get_class(actor_cfg.layer_norm.cls),
Expand All @@ -1155,7 +1155,7 @@ def build_agent(
input_dims=latent_state_size,
output_dim=critic_cfg.bins,
hidden_sizes=[critic_cfg.dense_units] * critic_cfg.mlp_layers,
activation=eval(critic_cfg.dense_act),
activation=hydra.utils.get_class(critic_cfg.dense_act),
layer_args={"bias": critic_ln_cls == nn.Identity},
flatten_dim=None,
norm_layer=critic_ln_cls,
Expand Down
Loading
Loading