gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-02-15 05:25:55 +01:00
Code Issues Projects Releases Wiki Activity

Removed tensorflow specific code in presets (#59)

Browse Source 
* Add generic layer specification for using in presets

* Modify presets to use the generic scheme
This commit is contained in:
Sina Afrooze
2018-11-06 07:39:29 -08:00
committed by Gal Leibovich
parent 811152126c
commit 93571306c3
19 changed files with 233 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
rl_coach/presets/Atari_NStepQ.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.n_step_q_agent import NStepQAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Conv2d, Dense
from rl_coach.architectures.layers import Conv2d, Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/BitFlip_DQN.py
View File

@@ -1,6 +1,6 @@
from rl_coach.agents.dqn_agent import DQNAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, EmbedderScheme, \
PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps

2
rl_coach/presets/BitFlip_DQN_HER.py
View File

@@ -1,6 +1,6 @@
from rl_coach.agents.dqn_agent import DQNAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, EmbedderScheme, \
PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps

33
rl_coach/presets/CARLA_CIL.py
View File

@@ -1,7 +1,6 @@
import os
import numpy as np
import tensorflow as tf
# make sure you have $CARLA_ROOT/PythonClient in your PYTHONPATH
from carla.driving_benchmark.experiment_suites import CoRL2017
from rl_coach.logger import screen
@@ -10,7 +9,7 @@ from rl_coach.agents.cil_agent import CILAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.head_parameters import RegressionHeadParameters
from rl_coach.architectures.middleware_parameters import FCMiddlewareParameters
from rl_coach.architectures.tensorflow_components.layers import Conv2d, Dense, BatchnormActivationDropout
from rl_coach.architectures.layers import Conv2d, Dense, BatchnormActivationDropout
from rl_coach.base_parameters import VisualizationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.carla_environment import CarlaEnvironmentParameters
@@ -46,34 +45,34 @@ agent_params.network_wrappers['main'].input_embedders_parameters = {
'CameraRGB': InputEmbedderParameters(
scheme=[
Conv2d(32, 5, 2),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(32, 3, 1),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(64, 3, 2),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(64, 3, 1),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(128, 3, 2),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(128, 3, 1),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(256, 3, 1),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Conv2d(256, 3, 1),
BatchnormActivationDropout(batchnorm=True, activation_function=tf.tanh),
BatchnormActivationDropout(batchnorm=True, activation_function='tanh'),
Dense(512),
BatchnormActivationDropout(activation_function=tf.tanh, dropout_rate=0.3),
BatchnormActivationDropout(activation_function='tanh', dropout_rate=0.3),
Dense(512),
BatchnormActivationDropout(activation_function=tf.tanh, dropout_rate=0.3)
BatchnormActivationDropout(activation_function='tanh', dropout_rate=0.3)
],
activation_function='none' # we define the activation function for each layer explicitly
),
'measurements': InputEmbedderParameters(
scheme=[
Dense(128),
BatchnormActivationDropout(activation_function=tf.tanh, dropout_rate=0.5),
BatchnormActivationDropout(activation_function='tanh', dropout_rate=0.5),
Dense(128),
BatchnormActivationDropout(activation_function=tf.tanh, dropout_rate=0.5)
BatchnormActivationDropout(activation_function='tanh', dropout_rate=0.5)
],
activation_function='none' # we define the activation function for each layer explicitly
)
@@ -84,7 +83,7 @@ agent_params.network_wrappers['main'].middleware_parameters = \
FCMiddlewareParameters(
scheme=[
Dense(512),
BatchnormActivationDropout(activation_function=tf.tanh, dropout_rate=0.5)
BatchnormActivationDropout(activation_function='tanh', dropout_rate=0.5)
],
activation_function='none'
)
@@ -94,9 +93,9 @@ agent_params.network_wrappers['main'].heads_parameters = [
RegressionHeadParameters(
scheme=[
Dense(256),
BatchnormActivationDropout(activation_function=tf.tanh, dropout_rate=0.5),
BatchnormActivationDropout(activation_function='tanh', dropout_rate=0.5),
Dense(256),
BatchnormActivationDropout(activation_function=tf.tanh)
BatchnormActivationDropout(activation_function='tanh')
],
num_output_head_copies=4 # follow lane, left, right, straight
)

2
rl_coach/presets/CartPole_PPO.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.clipped_ppo_agent import ClippedPPOAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters, DistributedCoachSynchronizationType
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.environments.gym_environment import GymVectorEnvironment, mujoco_v2

2
rl_coach/presets/ControlSuite_DDPG.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.ddpg_agent import DDPGAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, EmbedderScheme, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.control_suite_environment import ControlSuiteEnvironmentParameters, control_suite_envs

2
rl_coach/presets/Fetch_DDPG_HER_baselines.py
View File

@@ -1,7 +1,7 @@
from rl_coach.agents.ddpg_agent import DDPGAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.middleware_parameters import FCMiddlewareParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, EmbedderScheme, PresetValidationParameters
from rl_coach.core_types import EnvironmentEpisodes, EnvironmentSteps, TrainingSteps
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/Mujoco_A3C_LSTM.py
View File

@@ -1,7 +1,7 @@
from rl_coach.agents.actor_critic_agent import ActorCriticAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.middleware_parameters import LSTMMiddlewareParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, MiddlewareScheme, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/Mujoco_ClippedPPO.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.clipped_ppo_agent import ClippedPPOAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/Mujoco_DDPG.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.ddpg_agent import DDPGAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters, EmbedderScheme
from rl_coach.core_types import EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/Mujoco_NAF.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.naf_agent import NAFAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, GradientClippingMethod
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/Mujoco_PPO.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.ppo_agent import PPOAgentParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection

2
rl_coach/presets/Pendulum_HAC.py
View File

@@ -2,7 +2,7 @@ import numpy as np
from rl_coach.agents.hac_ddpg_agent import HACDDPGAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.architectures.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, EmbeddingMergerType, EmbedderScheme
from rl_coach.core_types import EnvironmentEpisodes, EnvironmentSteps, TrainingSteps
from rl_coach.environments.gym_environment import GymVectorEnvironment