h2o_sonar.lib.container package

Submodules

h2o_sonar.lib.container.explainer_container module

class h2o_sonar.lib.container.explainer_container.AsyncLocalContainer(max_workers: int | None = 3)

Bases: ExplainerContainer

Local container which runs interpretations asynchronously.

OPT_PER_INTERPRETATION_LOGGER: bool = True

TYPE_ID = 'ASYNC_LOCAL_EXPLAINER_CONTAINER'

property executor

gc()

Free system resources:

• shutdowns process pool(s)

• runs garbage collector

• clears temporary files

run_interpretation(dataset: str | Path | ExplainableDataset | Frame | DataFrame | hmli.H2OFrame, model: str | Path | ExplainableModel | Any, models: List[str | Path | ExplainableModel | Any], target_col: str, explainers: List[str | ExplainerToRun] | None = None, explainer_keywords: List[str] | None = None, validset: str | Path | ExplainableDataset | Frame | DataFrame | hmli.H2OFrame | None = None, testset: str | Path | ExplainableDataset | Frame | DataFrame | hmli.H2OFrame | None = None, use_raw_features: bool = True, used_features: [List] = None, weight_col: str = '', prediction_col: str = '', drop_cols: List | None = None, sample_num_rows: int | None = None, sampler: DatasetSampler | None = None, results_location: str | Path | Dict | None = None, results_formats: List[str] | None = None, progress_callback: AbstractProgressCallbackContext | None = None, run_asynchronously: bool = False, run_explainers_in_parallel: bool = False, key: str = '', extra_params: List | None = None) → Interpretation

setup(results_location: str | Any = '', persistence_api: PersistenceApi | None = None, persistence_type: PersistenceType | None = PersistenceType.file_system, logger=None, log_level: int | None = None)

Setup explainer container.

Parameters:

results_locationUnion[str, Any]

Optional directory where can container (interpretation, explainers, …) store data. If not specified, then in-memory store is used and logging to stdout/stderr is used.

persistence_apiOptional[p10s.PersistenceApi]

Instance of the persistence API allowing to create various persistence types (like file-system or DB)

persistence_typeOptional[p10s.Persistence]

Persistence type to be used by this container - default is the file-system persistence with the root in the current directory.

loggerOptional[loggers.SonarLogger]

Optional custom logger.

log_levelOptional[int]

Interpretation log level.

class h2o_sonar.lib.container.explainer_container.ContainerRegistry(singleton_create_key)

Bases: object

Explainer container factory and registry class is able to instantiate containers using symbolic names.

create_container(container_type_id)

Get instance of the container for given container type.

get_container_types() → List

list_containers() → Dict

classmethod register_container_type(container_type: Type[ExplainerContainer], container_type_id: str)

classmethod registry()

resolve_container(container: str | ExplainerContainer | None = None)

class h2o_sonar.lib.container.explainer_container.DependencyTreeNode(explainer_class, *children)

Bases: object

add_child(node)

get_node_by_id(eid) → DependencyTreeNode | None

str_lines()

to_list(result: List, queue_result: Dict, parent_queue: int = -1) → Tuple[List, Dict]

class h2o_sonar.lib.container.explainer_container.ExplainerContainer

Bases: ABC

Explainer container interface provides execution runtime independent APIs and functions which enable explainers to run “anywhere” via explainer container implementations for various runtimes.

Explainer gets reference to explainer container on its instantiation. Explainer container in turn contains references to public dataset, model and persistence APIs.

Explainer container can be instantiated with various combinations of mock/real/DB-based/… model/dataset/persistence APIs.

TYPE_ID = ''

delete_interpretation(mli_key: str)

property explainers_registry

Registry of explainers registered within the container.

gc()

Free system resources:

• shutdowns process pool(s)

• runs garbage collector

• clears temporary files

get_explainer(explainer_id) → ExplainerDescriptor

get_explainer_class(explainer_id) → Type[Explainer]

get_explainer_job_keys_by_id()

get_explainer_job_statuses(mli_key: str)

Get explainer job status for every explainer which was run within the interpretation.

get_explainer_local_result()

get_explainer_log_path()

get_explainer_metadata()

get_explainer_result_path()

get_explainer_snapshot_path()

get_interpretation_params(mli_key: str)

get_interpretation_status(mli_key: str)

Did interpretation succeeded and failed?

hot_deploy_explainer(explainer_descriptor: str)

Import custom BYOE recipe explainer class from given Python module. It is expected that the module is installed and/or on PYTHONPATH.

Parameters:

explainer_descriptorstr

Explainer descriptor with format ... for example ....

Returns:

Type[Explainer]

New explainer type.

list_explainers(experiment_types: List[str] | None = None, explanation_scopes: List[str] | None = None, model_meta: ExplainableModelMeta | None = None, keywords: List[str] | None = None, explainer_filter: List[FilterEntry] | None = None) → List[ExplainerDescriptor]

List and filter explainers.

Returns:

List[ExplainerDescriptor]

List of explainer descriptors which match the filter.

list_interpretations(username: str = 'h2o-sonar') → List[str]

load_interpretations(interpretation_key: str, username: str = 'h2o-sonar') → Interpretation

register_configured_explainers()

Register custom BYOE recipe explainers which are configured in H2O Eval Studio configuration.

register_explainer(explainer_class: Type[Explainer], explainer_id: str = '', extra_params: Dict | None = None)

run_interpretation(dataset: str | Path | ExplainableDataset | Frame | DataFrame | hmli.H2OFrame, model: str | Path | ExplainableModel | Any, models: List[str | Path | ExplainableModel | Any], target_col: str, explainers: List[str | ExplainerToRun] | None = None, explainer_keywords: List[str] | None = None, validset: str | Path | ExplainableDataset | ResourceHandle | Frame | DataFrame | ExplainableDatasetHandle | hmli.H2OFrame | None = None, testset: str | Path | ExplainableDataset | ResourceHandle | Frame | DataFrame | ExplainableDatasetHandle | hmli.H2OFrame | None = None, use_raw_features: bool = True, used_features: [List] = None, weight_col: str = '', prediction_col: str = '', drop_cols: List | None = None, sample_num_rows: int | None = 0, sampler: DatasetSampler | None = None, results_location: str | Path | Dict | None = None, results_formats: List[str] | None = None, progress_callback: AbstractProgressCallbackContext | None = None, run_asynchronously: bool = False, run_explainers_in_parallel: bool = False, key: str = '', logger: SonarLogger | None = None, extra_params: List | None = None) → Interpretation

abstract setup(results_location: str | Any = '', persistence_api: PersistenceApi | None = None, persistence_type: PersistenceType | None = PersistenceType.file_system, logger: SonarLogger | None = None, log_level: int | None = None)

Setup explainer container.

Parameters:

results_locationUnion[str, Any]

Optional directory where can container (interpretation, explainers, …) store data. If not specified, then in-memory store is used and logging to stdout/stderr is used.

persistence_apiOptional[p10s.PersistenceApi]

Instance of the persistence API allowing to create various persistence types (like file-system or DB)

persistence_typeOptional[p10s.Persistence]

Persistence type to be used by this container - default is the file-system persistence with the root in the current directory.

loggerOptional[loggers.SonarLogger]

Optional custom logger.

log_levelOptional[int]

Interpretation log level.

unregister_explainer(explainer_id: str, extra_params: Dict | None = None) → str

update_explainer_result()

class h2o_sonar.lib.container.explainer_container.ExplainerExecutor

Bases: ABC

Executor of the interpretation and its explainers.

class h2o_sonar.lib.container.explainer_container.LocalExplainerContainer

Bases: ExplainerContainer

Local explainer container enables explainers to run on local machine.

TYPE_ID = 'LOCAL_EXPLAINER_CONTAINER'

setup(results_location: str | Any = '', persistence_api: PersistenceApi | None = None, persistence_type: PersistenceType | None = PersistenceType.file_system, logger: SonarLogger | None = None, log_level: int | None = None)

Setup explainer container.

Parameters:

results_locationUnion[str, Any]

Optional directory where can container (interpretation, explainers, …) store data. If not specified, then in-memory store is used and logging to stdout/stderr is used.

persistence_apiOptional[p10s.PersistenceApi]

Instance of the persistence API allowing to create various persistence types (like file-system or DB)

persistence_typeOptional[p10s.Persistence]

Persistence type to be used by this container - default is the file-system persistence with the root in the current directory.

loggerOptional[loggers.SonarLogger]

Optional custom logger.

log_levelOptional[int]

Interpretation log level.

class h2o_sonar.lib.container.explainer_container.ParallelExplainerExecutor

Bases: ExplainerExecutor

Run interpretation explainers in parallel.

class PriorityQueues

Bases: object

Explainer executor priority queues.

HIGH_PRIORITY = 1

LOW_PRIORITY = 3

MEDIUM_PRIORITY = 2

classmethod ensure_valid_queue_plan(execution_plan: List, queue_plan: Dict, container_logger)

In 3 priority queues design, task from a higher priority queue cannot depend on task from a lower priority queue.

Parameters:

execution_planList

Execution plan which gives sequence of explainers to be executed.

queue_planDict

Queue plan to be verified (and eventually fixed).

container_logger

Container logger.

classmethod get_queue_for_explainer(explainer_keywords: List) → int

Get execution queue (high priority, medium priority, low priority) for explainer.

Parameters:

explainer_keywordsList

Explainer keywords where it requests the queue

Returns:

Priority queue ID.

class h2o_sonar.lib.container.explainer_container.SequentialExplainerExecutor(container: ExplainerContainer, explainers_registry: ExplainerRegistry, log_level, container_logger)

Bases: ExplainerExecutor

Run interpretation explainers sequentially.

This is “for loop style” explainers execution executor - no multithreading, no multiprocessing, no async - explainers are executed sequentially within the same process. This is straightforward, deadlock/race conditions free and system resources considerate approach.

run(interpretation: Interpretation) → Interpretation

h2o_sonar.lib.container.explainer_container.register_ootb_explainers(container: ExplainerContainer)

Module contents