max_num_steps=1000000,
logging_dir="half_cheetah/hierarchy_sac/",
num_hierarchy_levels=2,
time_skip=10,
hidden_size=256,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
initial_alpha=1.0,
policy_learning_rate=0.0003,
qf_learning_rate=0.0003,
tau=0.005,
batch_size=256,
max_path_length=1000,
num_workers=10,
num_warm_up_steps=100000,
num_steps_per_epoch=1000,
num_steps_per_eval=10000,
num_epochs_per_eval=10,
num_epochs=10000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in hierarchy_sac_variant.keys()])
# launch the experiment using ray
launch_local(
hierarchy_sac,
variant,
HalfCheetahEnv,
num_seeds=3)
from multiarchy.baselines.ddpg import ddpg, ddpg_variant
from multiarchy.envs.pointmass_env import PointmassEnv
if __name__ == "__main__":
# parameters for the learning experiment
variant = dict(max_num_steps=10000,
logging_dir="pointmass/ddpg/",
hidden_size=400,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
policy_learning_rate=0.0003,
qf_learning_rate=0.0003,
tau=0.005,
exploration_noise_std=0.1,
batch_size=256,
max_path_length=10,
num_workers=2,
num_warm_up_steps=100,
num_steps_per_epoch=10,
num_steps_per_eval=100,
num_epochs_per_eval=10,
num_epochs=10000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in ddpg_variant.keys()])
# launch the experiment using ray
launch_local(ddpg, variant, PointmassEnv, num_seeds=2)
from multiarchy.baselines.ppo import ppo, ppo_variant
from gym.envs.mujoco.half_cheetah import HalfCheetahEnv
if __name__ == "__main__":
# parameters for the learning experiment
variant = dict(max_path_length=1000,
max_num_paths=1000,
logging_dir="half_cheetah/ppo/",
hidden_size=400,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
epsilon=0.1,
lamb=0.95,
off_policy_updates=10,
critic_updates=32,
policy_learning_rate=0.0001,
vf_learning_rate=0.001,
num_workers=10,
num_steps_per_epoch=10000,
num_steps_per_eval=10000,
num_epochs_per_eval=1,
num_epochs=1000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in ppo_variant.keys()])
# launch the experiment using ray
launch_local(ppo, variant, HalfCheetahEnv, num_seeds=3)
variant = dict(
max_path_length=500,
max_num_paths=1000,
logging_dir="hopper_test2/ppo3/",
hidden_size=400,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
epsilon=0.1,
lamb=0.95,
off_policy_updates=10,
critic_updates=32,
policy_learning_rate=0.0001,
vf_learning_rate=0.001,
exploration_noise_std=0.5,
num_workers=10,
num_steps_per_epoch=5000,
num_steps_per_eval=5000,
num_epochs_per_eval=10,
num_epochs=1000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in ppo_variant.keys()])
# launch the experiment using ray
launch_local(
ppo,
variant,
HopperEnv,
num_seeds=1)
# parameters for the learning experiment
variant = dict(
max_num_steps=1000000,
logging_dir="humanoid/sac/",
hidden_size=400,
num_hidden_layers=2,
reward_scale=0.01,
discount=0.99,
initial_alpha=0.01,
lr=0.0003,
tau=0.005,
batch_size=256,
max_path_length=1000,
num_workers=2,
num_warm_up_steps=10000,
num_steps_per_epoch=1000,
num_steps_per_eval=10000,
num_epochs_per_eval=10,
num_epochs=10000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in sac_variant.keys()])
# launch the experiment using ray
launch_local(
sac,
variant,
HumanoidEnv,
num_seeds=1)
from multiarchy.baselines.ddpg import ddpg, ddpg_variant
from gym.envs.mujoco.hopper import HopperEnv
if __name__ == "__main__":
# parameters for the learning experiment
variant = dict(max_num_steps=1000000,
logging_dir="hopper_test2/ddpg/",
hidden_size=400,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
policy_learning_rate=0.0001,
qf_learning_rate=0.001,
tau=0.005,
exploration_noise_std=0.2,
batch_size=256,
max_path_length=500,
num_workers=2,
num_warm_up_steps=5000,
num_steps_per_epoch=500,
num_steps_per_eval=5000,
num_epochs_per_eval=10,
num_epochs=10000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in ddpg_variant.keys()])
# launch the experiment using ray
launch_local(ddpg, variant, HopperEnv, num_seeds=1)
"""Author: Brandon Trabucco, Copyright 2019, MIT License"""
from multiarchy.launch import launch_local
from multiarchy.baselines.policy_gradient import policy_gradient, policy_gradient_variant
from gym.envs.mujoco.hopper import HopperEnv
if __name__ == "__main__":
# parameters for the learning experiment
variant = dict(max_path_length=500,
max_num_paths=1000,
logging_dir="hopper_test2/pg/",
hidden_size=400,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
policy_learning_rate=0.0001,
exploration_noise_std=0.5,
num_workers=10,
num_steps_per_epoch=5000,
num_steps_per_eval=5000,
num_epochs_per_eval=10,
num_epochs=1000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in policy_gradient_variant.keys()])
# launch the experiment using ray
launch_local(policy_gradient, variant, HopperEnv, num_seeds=1)
from gym.envs.mujoco.hopper import HopperEnv
if __name__ == "__main__":
# parameters for the learning experiment
variant = dict(max_num_steps=1000000,
logging_dir="hopper/sac/",
hidden_size=256,
num_hidden_layers=2,
reward_scale=1.0,
discount=0.99,
initial_alpha=0.005,
policy_learning_rate=0.0003,
qf_learning_rate=0.0003,
tau=0.005,
batch_size=256,
max_path_length=1000,
num_workers=10,
num_warm_up_steps=1000,
num_steps_per_epoch=1000,
num_steps_per_eval=50000,
num_steps_per_gradient=1,
num_epochs_per_eval=10,
num_epochs=10000)
# make sure that all the right parameters are here
assert all([x in variant.keys() for x in sac_variant.keys()])
# launch the experiment using ray
launch_local(sac, variant, HopperEnv, num_seeds=5)