LIME

ExpyBox.lime()

Create dialog for lime

Returns

None

Method parameters

  • Number of features

    Maximum number of features present in explanation

  • Number of samples

    Size of the neighborhood to learn the surrogate linear model

  • Kernel width

    Kernel width for the exponential kernel. Actual value used will be the inputted value * sqrt(train_data.shape[1]).

  • Feature selection

    Feature selection method for choosing the best features for surrogate model. There are following options:

    • forward_selection: iteratively add features to the model (costly when num_features is high)

    • highest_weights: selects the features that have the highest product of absolute weight * original data point when learning with all the features

    • lasso_path: choose features based on the lasso regularization path

    • none: use all features, ignore Number of features option

    • auto: use forward_selection if Number of features <= 6, and highest_weights otherwise

  • Discretize continuous

    Whether to discretize all non-categorical features

  • Discretizer

    Which discretizer to use when discretizing continuous features. Only matters if discretize continuous is True.

  • Distance metric

    What distance metric to use for calculating weights of perturbed instances.

    It’s used as an distance_metric argument for sklearn.metrics.pairwise_distances(). Documentation of the function (with options for metric): sklearn.metrics.pairwise_distances

  • Variable with model regressor

    If you want to use a different regressor than Ridge regressor you can specify a variable in provided kernel_globals dictionary with the regressor.

    It must have model_regressor.coef_ and “sample_weight” as a parameter to model_regressor.fit()