Skip to main content

hydrogym.nek.configs

Configuration of runner and NEK @yuningw

Runner Objects​

@dataclass
class Runner()

RL_algorithm​

Policy to use

nb_episodes​

The total amount of Episodes is n_episode = nb_interactions * nb_episodes

nb_warmup_episodes​

Only for transfer learning, the number of episodes to warm up the critic

train_steps​

Off-policy only: Steps to train the model, N-train = n_episode // train_steps

ckpt_int​

Frequency of saving policy

n_epochs​

Setup of Pol

seed​

If -5000 -> None

random_init​

-1==No Shuffle, use the No.init = RANK; -2==NOT Cover the current rs8

rank​

NAME of test case

rew_mode​

We let all subdomain share the same reward

normalize_input​

Between None, utau, std

npl_state​

Should be consistent with SIZE

Simulation Objects​

@dataclass
class Simulation()

nproc​

ranks for running

TOTCTRL​

Be consistent with SIZE

znmf_avg​

1==Open

ys_bdf​

Volume for Body-Force

amp_bdf​

Amplitude for Body-Force

lx1​

Polynomial Order

Lz​

Size of domain

Nz​

No. Spectral Elements

writeControl​

runTime

PARFWRITE​

Do we write runtime parameter file

PARFNAME​

Runtime parameter file name for output (without .par)

LOGLEVEL​

Logging threshold for toolboxes

WALLTIME​

Simulation wall time

READCHKPT​

Restat from checkpoint

CHKPFNUMBER​

Restart file number

CHKPINTERVAL​

Checkpiont saving frequency (number of time steps)