Skip to content

plkit.optuna

module

plkit.optuna

</>

Optuna wrapper for plkit

Classes
  • OptunaSuggest — Optuna suggests for configuration items</>
  • Optuna — The class uses optuna to automate hyperparameter tuning</>
class

plkit.optuna.OptunaSuggest(default, suggtype, *args, **kwargs)

</>

Optuna suggests for configuration items

Parameters
  • default (any) — The default value, which the value will be collapsed to when optuna is opted out. So that you don't have to change your code if you don't run optuna.
  • suggtype (str) — The type of suggestion For example, cat refers to trial.suggest_categorical The mappings are: 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'
  • *args — The args used in trial.suggest_xxx(name, *args, **kwargs)
  • **kwargs — The kwargs used in trial.suggest_xxx(name, *args, **kwargs)
Attributes
  • args — *args from Args
  • default (any) — The default from Args
  • kwargs — **kwargs from Args
  • suggfunc (str) — The transformed suggestion name according to suggtype
Methods
  • suggest(name, trial) (Any) — Get the suggested value</>
method

suggest(name, trial)

</>

Get the suggested value

This is used in Optuna class, you don't have to call this

Parameters
  • name (str) — The name of the parameter
  • trial (optuna.Trial) — The trial to get the suggested value from
Returns (Any)

The suggested value

class

plkit.optuna.Optuna(on, n_trials, **kwargs)

</>

The class uses optuna to automate hyperparameter tuning

Example 
>>> from plkit import Data, Module, Optuna
>>> class MyData(Data):
>>>     ...
>>> class MyModel(Module):
>>>     ...
>>> class MyOptuna(Optuna):
>>>     def suggests(self, config):
>>>         ...
>>>         return new_config
>>> optuna = MyOptuna('val_loss', 100)
>>> optuna.run(config, model_class, data_class)
Parameters
  • on (str) — On which value to optimize. Should be one of the keys of dictionary that is returned from validation_epoch_end. val_acc or val_loss for example.
  • n_trials (int) — Number of trials
  • **kwargs — Other keyword arguments for optuna.create_study
Attributes
  • best_model — Get the model from best trainer</>
  • best_params — The best parameters from the study</>
  • best_trainer — Get the best trainer</>
  • best_trial — The best trial from the study</>
  • n_trials (int) — n_trials from Args
  • on (str) — on from Args
  • study (optuna.Study) — study object created from kwargs
  • trainers (list) — list of trainers to keep track of the best one
  • trials — The trials</>
Methods
  • run(config, data_class, model_class, **kwargs) — Run the optimization</>
  • run(config, data_class, model_class, **kwargs) — Run the optimization</>
  • suggests(trial, conf) (dict) — Collect the hyperparameters from the trial suggestions if any configuration item is an OptunaSuggest object</>
method

suggests(trial, conf)

</>

Collect the hyperparameters from the trial suggestions if any configuration item is an OptunaSuggest object

Parameters
  • trial (optuna.Trial) — the trial object
  • conf (dict) — The configuration dictionary
Returns (dict)

A dictionary of suggested parameters

method

run(config, data_class, model_class, **kwargs)

</>

Run the optimization

The optimization is running on fit of the trainer. If test data is provided. Test will be performed as well.

Parameters
  • config (dict) — The configuation dictionary
  • data_class (class) — The data class subclassed from plkit.Data Note that this is the class itself, not the instantized object.
  • model_class (class) — The data class subclassed from plkit.Module Note that this is the class itself, not the instantized object.
  • **kwargs — Other arguments for study.optimize other than func and n_trials. See: https://optuna.readthedocs.io/en/stable/reference/generated/optuna.study.Study.html#optuna.study.Study.optimize
method

run(config, data_class, model_class, **kwargs)

</>

Run the optimization

The optimization is running on fit of the trainer. If test data is provided. Test will be performed as well.

Parameters
  • config (dict) — The configuation dictionary
  • data_class (class) — The data class subclassed from plkit.Data Note that this is the class itself, not the instantized object.
  • model_class (class) — The data class subclassed from plkit.Module Note that this is the class itself, not the instantized object.
  • **kwargs — Other arguments for study.optimize other than func and n_trials. See: https://optuna.readthedocs.io/en/stable/reference/generated/optuna.study.Study.html#optuna.study.Study.optimize