Optuna integration

plkit integrates the optuna framework to tune the hyperparameters. You don't have to modify your code to much to enable it by following our best practice.

To enable optuna, you just need to pass an optuna object to plkit.run or runner.run method.

To initialize an optuna object, you need to pass in following arguments:

• on: telling plkit which metric to optimize on. You need to log that metric inside validation_epoch_end for the object to fetch it.
• n_trials: How many trials you want to run
• **kwargs: Other arguments for optuna.create_study (see optuna.study.create_study).

OptunaSuggest

Defines how the parameters should be tuning. You can specify an OptunaSuggest object to a parameter in configuration like this:

configuration = {
    #                             default, type, low, high
    'learning_rate': OptunaSuggest(1e-3, 'float', 1e-5, 5e-2)
}

This way, yon don't have to modify the configuration when you want to enable or disable optuna integration.

When optuna is enabled, a value will be generated for learning_rate using trial.suggest_float('learning_rate', 1e-5, 5e-2) and the default value is ignored. While it's disabled, it will fall back to the default value.

See all avaiable suggest types here and corresponding names (aliases) for the types as the second argument of OptunaSuggest

Name/Alias	optuna.trial.Trial.suggest_xxx
cat	suggest_categorical
categorical	suggest_categorical
distuni	suggest_discrete_uniform
dist_uni	suggest_discrete_uniform
discrete_uniform	suggest_discrete_uniform
float	suggest_float
int	suggest_int
loguni	suggest_loguniform
log_uni	suggest_loguniform
uni	suggest_uniform