logml.model_search.provider

Models provider interface

Classes

ModelProvider([output_structure, config, ...])

Class to save or acquire model instances whenever they are required, like feature importance module.

class logml.model_search.provider.ModelProvider(output_structure: Optional[logml.model_search.ModelSearchOutputStructure] = None, config: Optional[logml.configuration.modeling.ModelSearchSection] = None, logger=None, auto_load=True, random_state=None)

Bases: object

Class to save or acquire model instances whenever they are required, like feature importance module.

If dumped models data is not available, default models are created.

dump_selection_result(result: Optional[logml.model_search.common.ModelSelectionResult] = None): Dump selection result.

dump_model_hpo_result(hpo_result: Optional[logml.model_search.hp_optimizer.HpoResult] = None, ds_name: Optional[str] = None): Dump model HPO result and trials.

dump_model_result(model_result: Optional[logml.model_search.common.ModelEvaluationData] = None, ds_name: Optional[str] = None, model: Optional[logml.models.base.BaseModel] = None): Dump model training and evaluation result.

load_model_evaluation_result(model_name: str, ds_name: str) → Optional[logml.model_search.common.ModelEvaluationData]: Load a given model’s training and evaluation result for a given dataset.

load_model_instance(model_name: Optional[str] = None, ds_name: Optional[str] = None) → Optional[logml.models.base.BaseModel]: Load pickled model instance

load_model_hpo_result(model_id: Optional[str] = None, ds_name: Optional[str] = None) → Optional[logml.model_search.hp_optimizer.ModelHpoResult]: Load model HPO result and trials.

load_hpo_result() → Optional[List[logml.model_search.hp_optimizer.ModelHpoResult]]: Load all models HPO result

load_selection_result(): Load previously dumped summary

get_model_class(model_label: Optional[str] = None) → Type[logml.models.base.BaseModel]: Returns class for new models creation by caller.

get_configured_models() → Optional[list]: Returns configured models, disregarding selected ones.

create_model_instance(model_name)

get_model_data(model_name: str, check_nonselected=False): Gets model params, checking from concrete instance to most general configs up.

get_selected_models() → Optional[list]: Returns models parameters picked by models selection process.

sort_selection_result(): Orders existing models by pvalue and loss.