facet.explanation.parallel.ExplainerQueue#

class facet.explanation.parallel.ExplainerQueue(explainer, X, y=None, *, interactions, max_job_size, **kwargs)[source]#

A queue splitting a data set to be explained into multiple jobs.

Bases

Metaclasses

ABCMeta

Parameters

explainer (BaseExplainer) – the SHAP explainer to use
X (Union[ndarray[Any, dtype[Any]], DataFrame, Pool]) – the feature values of the observations to be explained
y (Union[ndarray[Any, dtype[Any]], Series, None]) – the target values of the observations to be explained
interactions (bool) – if False, calculate SHAP values; if True, calculate SHAP interaction values
max_job_size (int) – the maximum number of observations to allocate to each job
kwargs (Any) – additional arguments specific to the explanation method

Raises

NotImplementedError – if X is a Pool; this is currently not supported

Method summary

`aggregate`	Called by `JobRunner.run_queue()` to aggregate the results of all jobs once they have all been run.
`jobs`	Iterate the jobs in this queue.
`on_run`	See `pytools.parallelization.JobQueue.on_run()`

Attribute summary

`lock`	The lock used by class `JobRunner` to prevent parallel executions of the same queue
`explainer`	the SHAP explainer to use
`interactions`	if `False`, calculate SHAP values; otherwise, calculate SHAP interaction values
`X`	the feature values of the observations to be explained
`y`	the target values of the observations to be explained
`max_job_size`	the maximum number of observations to allocate to each job
`kwargs`	additional arguments specific to the explanation method

Definitions

aggregate(job_results)[source]#

Called by JobRunner.run_queue() to aggregate the results of all jobs once they have all been run.

Parameters: job_results (List[Union[ndarray[Any, dtype[Any]], List[ndarray[Any, dtype[Any]]]]]) – list of job results, ordered corresponding to the sequence of jobs generated by method jobs()
Return type: Union[ndarray[Any, dtype[Any]], List[ndarray[Any, dtype[Any]]]]
Returns: the aggregated result of running the queue

Iterate the jobs in this queue.

Return type: Iterable[Job[Union[ndarray[Any, dtype[Any]], List[ndarray[Any, dtype[Any]]]]]]
Returns: the jobs in this queue

X: numpy.ndarray[Any, numpy.dtype[Any]]#: the feature values of the observations to be explained

interactions: bool#: if False, calculate SHAP values; otherwise, calculate SHAP interaction values

kwargs: Dict[str, Any]#: additional arguments specific to the explanation method

max_job_size: int#: the maximum number of observations to allocate to each job

y: Optional[numpy.ndarray[Any, numpy.dtype[Any]]]#: the target values of the observations to be explained

facet.explanation.parallel.ExplainerJob

facet.explanation.parallel.ParallelExplainer