arthurai.explainability.explanation_packager.ExplanationPackager#

class arthurai.explainability.explanation_packager.ExplanationPackager(model, **kwargs)#

Bases: object

Manages the project source code and explainer for this model. Intended only for internal use. Used to create explainer object, package files and format config for the update enrichments API endpoint.

See model.update_enrichments() for more details on usage

Methods

`build_categorical_lookup_table`	Builds the lookup table that ArthurExplainer needs for encoding strings
`contains_file_fields`	Returns (onePresent, allPresent) onePresent is true if at least one file field was supplied AllPresent is true if all file fields are supplied
`copy_files`	Copy the files that are required for testing the docker image.
`create`	Create an explainer for this model.
`create_test_script`	Create a test script.
`get_request_config`	Generates config dict for explainer.
`get_request_files`	Returns files needed for request, for use with model.update_enrichments
`make_zip`	Create the zip file for the project directory.
`test`	Download the docker container, install the files, and make sure the predict function can be run.
`validate_requirements_add_sklearn`	Checks the user requirements file to make sure versions are correct.

Attributes

FILE_FIELDS

build_categorical_lookup_table()#

Builds the lookup table that ArthurExplainer needs for encoding strings

Return type: List[Optional[Dict[str, int]]]

contains_file_fields()#

Returns (onePresent, allPresent) onePresent is true if at least one file field was supplied AllPresent is true if all file fields are supplied

Return type: Tuple[bool, bool]

copy_files(example)#

Copy the files that are required for testing the docker image.

Param: example: the example to use in the test script
Return type: None

create()#

Create an explainer for this model.

Param: df: the training data for the explainer
Param: shap_nrows: number of rows to sample the dataframe passed in
Return type: None

create_test_script(example)#

Create a test script.

Param: example:the example to use in the test script
Return type: str
Returns: the script contents

get_request_config()#

Generates config dict for explainer. For use with model.update_enrichments

Return type: Dict[str, str]

get_request_files()#

Returns files needed for request, for use with model.update_enrichments

Return type: List[Any]

make_zip()#

Create the zip file for the project directory.

Return type: None

test()#

Download the docker container, install the files, and make sure the predict function can be run.

Return type: None

validate_requirements_add_sklearn()#

Checks the user requirements file to make sure versions are correct.

Also handles making sure scikit-learn version is pinned in user requirements file, if it is installed in local environment and not specified.

scikit-learn is a dependency of shap, and arthurai doesn’t require scikit-learn explicitly. This can lead to version A being installed and used locally when pickling explainer, then version B getting installed on model server as part of shap install. This can lead to pickle errors, since the pickled version of shap (client side) used a different version of sklearn than what is on model server.

Return type: List[str]