Best break_step value

[farzinaa]

Hi,
Any idea how one could optimize the value of break_step as it has a great impact on the final model. I tried to pass this through Optuna unsuccessfully.
Any help or reading material would be greatly appropriated. Thanks.

[farzinaa]

This seems to be working if anyone is interested:

def sample_params(trial:optuna.Trial,):
break_step = trial.suggest_categorical("break_step", [1e4, 1e5])
gamma = trial.suggest_categorical("gamma", [0.8, 0.9, 0.95, 0.98, 0.99])
learning_rate = trial.suggest_loguniform("learning_rate", 1e-6, 1e-3)
ERL_PARAMS = {
"break_step":break_step,
"gamma": gamma,
"learning_rate": learning_rate,
}
return ERL_PARAMS
def objective(trial:optuna.Trial):
hyperparameters = sample_params(trial)
ERL_PPO_PARAMS = dict(itertools.islice(hyperparameters.items(), 2, 3))
break_step = hyperparameters['break_step']
train(
drl_lib='elegantrl',
env=env,
model_name='ppo',
erl_params= ERL_PPO_PARAMS,
cwd='./papertrading_erl',
break_step = break_step
)
etc.

Also it looks like the break step is not all that relevant if you are using callbacks (LoggingCallback in the tutorial) when optimizing your optuna.