DeepScoringExplainer Class
Defines a scoring model based on DeepExplainer.
If the original explainer was using a SHAP DeepExplainer and no initialization data was passed in, the core of the original explainer will be reused. If the original explainer used another method or new initialization data was passed in under initialization_examples, a new explainer will be created.
If transformations was passed in on original_explainer, those transformations will be carried through to the scoring explainer, it will expect raw data, and by default importances will be returned for raw features. If feature_maps are passed in here (NOT intended to be used at the same time as transformations), the explainer will expected transformed data, and by default importances will be returned for transformed data. In either case, the output can be specified by setting get_raw explicitly to True or False on the explainer's explain method.
Initialize the DeepScoringExplainer.
If the original explainer was using a SHAP DeepExplainer and no initialization data was passed in, the core of the original explainer will be reused. If the original explainer used another method or new initialization data was passed in under initialization_examples, a new explainer will be created.
If transformations was passed in on original_explainer, those transformations will be carried through to the scoring explainer, it will expect raw data, and by default importances will be returned for raw features. If feature_maps are passed in here (NOT intended to be used at the same time as transformations), the explainer will expected transformed data, and by default importances will be returned for transformed data. In either case, the output can be specified by setting get_raw explicitly to True or False on the explainer's explain method.
- Inheritance
-
azureml.interpret.scoring.scoring_explainer._scoring_explainer.ScoringExplainerDeepScoringExplainer
Constructor
DeepScoringExplainer(original_explainer, initialization_examples=None, serializer=None, **kwargs)Parameters
| Name | Description |
|---|---|
|
original_explainer
Required
|
<xref:interpret_community.common.base_explainer.BaseExplainer>
The training time deep explainer originally used to explain the model. |
|
initialization_examples
|
A matrix of feature vector examples (# examples x # features) for initializing the explainer. Default value: None
|
|
serializer
|
Picklable custom serializer with save and load methods defined for model that is not serializable. The save method returns a dictionary state and load method returns the model. Default value: None
|
|
feature_maps
Required
|
A list of feature maps from raw to generated feature. The list can be numpy arrays or sparse matrices where each array entry (raw_index, generated_index) is the weight for each raw, generated feature pair. The other entries are set to zero. For a sequence of transformations [t1, t2, ..., tn] generating generated features from raw features, the list of feature maps correspond to the raw to generated maps in the same order as t1, t2, etc. If the overall raw to generated feature map from t1 to tn is available, then just that feature map in a single element list can be passed. |
|
raw_features
Required
|
Optional list of feature names for the raw features that can be specified if the original explainer computes the explanation on the engineered features. |
|
engineered_features
Required
|
Optional list of feature names for the engineered features that can be specified if the original explainer has transformations passed in and only computes the importances on the raw features. |
|
original_explainer
Required
|
<xref:interpret_community.common.base_explainer.BaseExplainer>
The training time deep explainer originally used to explain the model. |
|
initialization_examples
Required
|
A matrix of feature vector examples (# examples x # features) for initializing the explainer. |
|
serializer
Required
|
<xref:<xref:object with save method that returns a dictionary state and load method that returns the model.>>
Picklable custom serializer with save and load methods defined for model that is not serializable. |
|
feature_maps
Required
|
A list of feature maps from raw to generated feature. The list can be numpy arrays or sparse matrices where each array entry (raw_index, generated_index) is the weight for each raw, generated feature pair. The other entries are set to zero. For a sequence of transformations [t1, t2, ..., tn] generating generated features from raw features, the list of feature maps correspond to the raw to generated maps in the same order as t1, t2, etc. If the overall raw to generated feature map from t1 to tn is available, then just that feature map in a single element list can be passed. |
|
raw_features
Required
|
Optional list of feature names for the raw features that can be specified if the original explainer computes the explanation on the engineered features. |
|
engineered_features
Required
|
Optional list of feature names for the engineered features that can be specified if the original explainer has transformations passed in and only computes the importances on the raw features. |
Methods
| explain |
Use the DeepExplainer and deep model for scoring to get the feature importance values of data. |
explain
Use the DeepExplainer and deep model for scoring to get the feature importance values of data.
explain(evaluation_examples, get_raw=None)Parameters
| Name | Description |
|---|---|
|
evaluation_examples
Required
|
A matrix of feature vector examples (# examples x # features) on which to explain the model's output. |
|
get_raw
|
If True, importance values for raw features will be returned. If False, importance values for engineered features will be returned. If unspecified and transformations was passed into the original explainer, raw importances will be returned. If unspecified and feature_maps was passed into the scoring explainer, engineered importance values will be returned. Default value: None
|
Returns
| Type | Description |
|---|---|
|
For a model with a single output such as regression, this returns a matrix of feature importance values. For models with vector outputs this function returns a list of such matrices, one for each output. The dimension of this matrix is (# examples x # features). |