def test_agent():
from rtrl import Training, run
Sac_Test = partial(
Training,
epochs=3,
rounds=5,
steps=100,
Agent=partial(Agent,
memory_size=1000000,
start_training=256,
batchsize=4),
Env=partial(id="Pendulum-v0", real_time=0),
)
run(Sac_Test)
def test_agent_avenue():
from rtrl import Training, run
from rtrl.envs import AvenueEnv
Sac_Avenue_Test = partial(
Training,
epochs=3,
rounds=5,
steps=300,
Agent=partial(AvenueAgent,
device='cpu',
training_interval=4,
start_training=400),
Env=partial(AvenueEnv, real_time=0),
Test=partial(number=0), # laptop can't handle more than that
)
run(Sac_Avenue_Test)
os.mkdir(path)
save_json(partial_to_dict(run_cls), path + '/spec.json')
if not exists(path + '/stats'):
dump(pd.DataFrame(), path + '/stats')
for stats in iterate_episodes(run_cls, path + '/state'):
dump(
load(path + '/stats').append(stats, ignore_index=True),
path + '/stats') # concat with stats from previous episodes
# === specifications ===================================================================================================
TestTraining = partial(
Training,
epochs=3,
rounds=5,
steps=10,
Agent=partial(memory_size=1000000),
Env=partial(id="Pendulum-v0"),
)
SacTraining = partial(Training,
Agent=partial(rtrl.sac.Agent),
Env=partial(id="Pendulum-v0"),
Test=partial(number=4))
RtacTraining = partial(
SacTraining,
Agent=partial(rtrl.rtac.Agent),
Env=partial(real_time=True),
)
self.outputnorm.parameters(),
self.target_update)
return dict(
loss_actor=loss_actor.detach(),
loss_critic=loss_critic.detach(),
outputnorm_mean=float(self.outputnorm.mean),
outputnorm_std=float(self.outputnorm.std),
memory_size=len(self.memory),
)
AvenueAgent = partial(Agent,
entropy_scale=0.05,
lr=0.0002,
memory_size=500000,
batchsize=100,
training_interval=4,
start_training=10000,
Model=partial(rtrl.sac_models.ConvModel))
# === tests ============================================================================================================
def test_agent():
from rtrl import Training, run
Sac_Test = partial(
Training,
epochs=3,
rounds=5,
steps=100,
Agent=partial(Agent,
memory_size=1000000,
return dict(
loss_total=loss_total.detach(),
loss_critic=loss_critic.detach(),
loss_actor=loss_actor.detach(),
outputnorm_mean=float(self.outputnorm.mean),
outputnorm_std=float(self.outputnorm.std),
memory_size=len(self.memory),
# entropy_scale=self.entropy_scale
)
AvenueAgent = partial(Agent,
entropy_scale=0.05,
lr=0.0002,
memory_size=500000,
batchsize=100,
training_interval=4,
start_training=10000,
Model=partial(ConvDouble))
if __name__ == "__main__":
from rtrl import Training, run
from rtrl import rtac_models
Rtac_Test = partial(
Training,
epochs=3,
rounds=5,
steps=500,
Agent=partial(Agent,
device='cpu',
memory_size=1000000,
os.mkdir(path)
save_json(partial_to_dict(run_cls), path + '/spec.json')
if not exists(path + '/stats'):
dump(pd.DataFrame(), path + '/stats')
for stats in iterate_episodes(run_cls, path + '/state'):
dump(
load(path + '/stats').append(stats, ignore_index=True),
path + '/stats') # concat with stats from previous episodes
# === specifications ===================================================================================================
TestTraining = partial(
Training,
epochs=3,
rounds=5,
steps=10,
Agent=partial(memory_size=1000000),
Env=partial(id="Pendulum-v0"),
)
SacTraining = partial(Training,
Agent=partial(rtrl.sac.Agent),
Env=partial(id="Pendulum-v0"),
Test=partial(number=4))
RtacTraining = partial(
SacTraining,
Agent=partial(rtrl.rtac.Agent),
Env=partial(real_time=True),
)
return dict(
loss_total=loss_total.detach(),
loss_critic=loss_critic.detach(),
loss_actor=loss_actor.detach(),
outputnorm_mean=float(self.outputnorm.mean),
outputnorm_std=float(self.outputnorm.std),
memory_size=len(self.memory),
# entropy_scale=self.entropy_scale
)
AvenueAgent = partial(Agent,
entropy_scale=0.05,
lr=0.0002,
memory_size=500000,
batchsize=100,
training_interval=4,
start_training=10000,
Model=partial(ConvDouble))
if __name__ == "__main__":
from rtrl import Training, run
from rtrl import rtac_models
# Rtac_Test = partial(
# Training,
# epochs=3,
# rounds=5,
# steps=500,
# Agent=partial(Agent, device='cpu', memory_size=1000000, start_training=256, batchsize=4),
# Env=partial(id="Pendulum-v0", real_time=True),
# # Env=partial(id="HalfCheetah-v2", real_time=True),