logml.model_search
Classes
|
Defines output structure for Model Search endpoint. |
|
Files specific for (model, dataset) pair. |
- class logml.model_search.ModelSearchResultOutputStructure(artifacts_root: pathlib.Path)
Bases:
object
Files specific for (model, dataset) pair.
- MODEL_EVALUATION_FILENAME: str = 'evaluation.pickle'
- MODEL_INSTANCE_FILENAME: str = 'model.pickle'
- MODEL_HPO_FILENAME: str = 'hpo.pickle'
- property model_evaluation_file_path
Model evaluation result file path.
- property model_instance_file_path
Model instance file path.
- property model_hpo_result_file_path
Model hpo result file path.
- class logml.model_search.ModelSearchOutputStructure(*args, **kwargs)
Bases:
logml.common.OutputStructure
Defines output structure for Model Search endpoint. Structure:
- experiment_root/…
- /model_search/{problem_id}/
- selection_result.json * data for model selection result
(list of good models)
- /{model_id}__{dataset_hash} * data for specific model/ds:
evaluation.pickle * model params and metrics
hpo.pickle * hyperparams data
model.pickle * tuned model for later use
- MODEL_SEARCH: str = 'model_search'
- SUMMARY_FILENAME: str = 'selection_result.json'
- DELIMETER = '__'
- get_summary_file_path()
Get path to summary file
- get_model_folder_path(model_id: str, dataset_name: str) pathlib.Path
Returns path to a folder for specific model/dataset data.
- list_available_models(dataset_name: str) List[str]
For a given dataset hash returns a list of models with existing evaluation results.
- get_model_structure(model_id: str, dataset_name: str) logml.model_search.ModelSearchResultOutputStructure
Returns path to a folder for specific model/dataset data.
- artifacts_exist() bool
Check whether summary artifact exists.
Modules
Models provider interface |
|
Models training and selection |
|