Interpretation Parameters

The following parameters can be specified when starting a new interpretation using the Python API:

  • dataset : Union[str, Path, ExplainableDataset, datatable.Frame, Any]

    • Dataset source: explainable dataset instance, datatable frame, string (path to CSV, .jay or any other file type supported by datatable), handle to a remote dataset hosted by Driverless AI or a dictionary (used to construct frame).

  • model : Union[str, Path, ExplainableModel, Any]

    • Path to model (str, Path), explainable model (ExplainableModel) , handle to a remote model hosted by Driverless AI or an instance of 3rd party model (like Scikit) to interpret.

  • target_col : str

    • Target column name - must be valid dataset column name.

  • explainers : Optional[List[Union[str, commons.ExplainerToRun]]]

    • Explainer IDs to run within the interpretation or ExplainerToRun instances with explainer parameters. In case of None or empty list are run all compatible explainers.

  • explainer_keywords: Optional[List[str]]

    • Run compatible explainers which have given keyword (AND). This setting is used only in case that explainers parameter is empty list (or None).

  • validset : Optional[Union[src, Path, ExplainableDataset, datatable.Frame, Any]]

    • Optional validation dataset - sources might the same as in case of dataset.

  • testset : Optional[Union[src, Path, ExplainableDataset, datatable.Frame, Any]]

    • Optional test dataset - sources might the same as in case of dataset.

  • use_raw_features : bool

    • True to use original features, False to use transformed features.

  • used_features : Optional[List]

    • Optional parameter specifying features (dataset columns) used by the model. This parameter is used in case that an instance of the model (not ExplainableModel) is provided by the user - therefore ExplainableModel’s metadata are not available.

  • weight_col : str

    • Name of the weight column to be used by explainers.

  • prediction_col : str

    • Name of the predictions column - in case of 3rd party model (standalone MLI).

  • drop_cols : Optional[List]

    • List of the columns to drop from the interpretation i.e. columns names which should not be explained.

  • sample_num_rows : Optional[int]

    • Sample the dataset to given number of rows. By default, the dataset is sampled based on the RAM size (or to 25000 rows). Use 0 to disable sampling or use an integer value greater than 0 to sample to the specified number of rows.

  • sampler : Optional[DatasetSampler]

    Sampling method (implementation) to be used - see h2o_sonar.utils.sampling module (documentation) for available sampling methods. Use a sampler instance to use the specific sampling method.

  • container : Optional[Union[str, explainer_container.ExplainerContainer]]

    • Optional explainer container name (str) or container instance to be used to run the interpretation.

  • results_location : Optional[Union[str, pathlib.Path, Dict, Any]]

    • Where to store interpretation results - filesystem (path as string or Path), memory (dictionary) or DB. If None, then results are stored to the current directory.

  • persistence_type : persist.PersistenceType

    • Optional choice of the persistence type: file-system (default), in-memory or database. This option does not override persistence type in case that container is provided.

  • args_as_json_location : Optional[Union[str, pathlib.Path]]

    • Load all positional arguments and keyword arguments from JSon file. This is useful when input is generated, persisted, repeated and used from CLI (which doesn’t support all the options). IMPORTANT: if this argument is specified, then all other function parameters are ignored.

  • upload_to : Union[str, config.ConnectionConfig]

    • Upload the interpretation report to the H2O GPT Enterprise in order to talk to the report.

  • log_level : int

    • Optional container and explainers log level.

See also: