Interpretation Expert Settings

The following is a list of the Interpretation expert settings that are available when setting up a new interpretation from the MLI page. The name of each setting is preceded by its config.toml label. For info on explainer-specific expert settings, see Explainer (Recipes) Expert Settings.

Interpretation Expert Settings

MLI Tab

mli_sample

Sample All Explainers

Specify whether to perform the interpretation on a sample of the training data. By default, MLI will sample the training dataset if it is greater than 100k rows. (The equivalent config.toml setting is mli_sample_size.) This is enabled by default. Turn this toggle off to run MLI on the entire dataset.

mli_enable_mojo_scorer

Allow Use of MOJO Scoring Pipeline

Use this option to disable MOJO scoring pipeline. Scoring pipeline is chosen automatically (from MOJO and Python pipelines) by default. In case of certain models, MOJO vs. Python choice can impact pipeline performance and robustness.

mli_fast_approx

Speed up predictions with a fast approximation

Specify whether to speed up predictions with a fast approximation. When enabled, this setting can reduce the number of trees or cross-validation folds and ultimately reduce the time needed to complete interpretations. This setting is enabled by default.

mli_custom

Add to config.toml via TOML String

Use this input field to add to the Driverless AI server config.toml configuration file with TOML string.

MLI NLP Tab

mli_nlp_top_n

Number of Tokens Used for MLI NLP Explanations

Specify the number of tokens used for MLI NLP explanations. To use all available tokens, set this value to -1. By default, this value is set to 20.

mli_nlp_sample_limit

Sample Size for NLP Surrogate Models

Specify the maximum number of records used by MLI NLP explainers. The default value is 10000.

mli_nlp_min_df

Minimum Number of Documents in Which Token Has to Appear

Specify the minimum number of documents in which token has to appear. Use integer values to denote absolute counts and floating-point values to denote percentages. By default, this value is set to 3.

mli_nlp_max_df

Maximum Number of Documents in Which Token Has to Appear

Specify the maximum number of documents in which token has to appear. Use integer values to denote absolute counts and floating-point values to denote percentages. By default, this value is set to 3.

mli_nlp_min_ngram

Minimum Value in n-gram Range

Specify the minimum value in the n-gram range. The tokenizer generates all possible tokens in the range specified by mli_nlp_min_ngram and mli_nlp_max_ngram. By default, this value is set to 1.

mli_nlp_max_ngram

Maximum Value in n-gram Range

Specify the maximum value in the n-gram range. The tokenizer generates all possible tokens in the range specified by mli_nlp_min_ngram and mli_nlp_max_ngram. By default, this value is set to 1.

mli_nlp_min_token_mode

Mode Used to Choose N Tokens for MLI NLP

Specify the mode used to choose N tokens. Select from the following:

  • top - Chooses N top tokens

  • bottom - Chooses N bottom tokens

  • top-bottom - Chooses math.floor (N/2) top and math.ceil (N/2) bottom tokens

  • linspace - Chooses N evenly spaced tokens

mli_nlp_tokenizer_max_features

Number of Top Tokens to Use as Features (Token-based Feature Importance)

Specify the number of top tokens to use as features when building token-based feature importance. By default, this value is set to -1.

mli_nlp_loco_max_features

Number of Top Tokens to Use as Features (LOCO)

Specify the number of top tokens to use as features when computing text LOCO. By default, this value is set to -1.

mli_nlp_surrogate_tokens

Number of Top Tokens to Use as Features (Surrogate Model)

Specify the number of top tokens to use as features when building surrogate models. Note that this setting only applies to NLP models. By default, this value is set to 100.

mli_nlp_use_stop_words

Stop Words for MLI NLP

Specify whether to use stop words for MLI NLP. This setting is enabled by default.

mli_nlp_stop_words

List of Words to Filter Before Generating Text Tokens

Specify a list of words to filter out before generating text tokens, which are passed to MLI NLP LOCO and surrogate models (if enabled). Enter a custom list of stop words. For example, you can enter ['great', 'good'] to filter out the words great and good.

mli_nlp_append_to_english_stop_words

Append List of Custom Stop Words to Default Stop Words

Specify whether to append the list of stop words specified by mli_nlp_stop_words to the default list of stop words. This setting is disabled by default.

MLI Surrogate Models Tab

mli_lime_method

LIME Method

Select a LIME method of either K-LIME (default) or LIME-SUP.

  • K-LIME (default): creates one global surrogate GLM on the entire training data and also creates numerous local surrogate GLMs on samples formed from k-means clusters in the training data. The features used for k-means are selected from the Random Forest surrogate model’s variable importance. The number of features used for k-means is the minimum of the top 25% of variables from the Random Forest surrogate model’s variable importance and the max number of variables that can be used for k-means, which is set by the user in the config.toml setting for mli_max_number_cluster_vars. (Note, if the number of features in the dataset are less than or equal to 6, then all features are used for k-means clustering.) The previous setting can be turned off to use all features for k-means by setting use_all_columns_klime_kmeans in the config.toml file to true. All penalized GLM surrogates are trained to model the predictions of the Driverless AI model. The number of clusters for local explanations is chosen by a grid search in which the \(R2\) between the Driverless AI model predictions and all of the local K-LIME model predictions is maximized. The global and local linear model’s intercepts, coefficients, \(R2\) values, accuracy, and predictions can all be used to debug and develop explanations for the Driverless AI model’s behavior.

  • LIME-SUP: explains local regions of the trained Driverless AI model in terms of the original variables. Local regions are defined by each leaf node path of the decision tree surrogate model instead of simulated, perturbed observation samples - as in the original LIME. For each local region, a local GLM model is trained on the original inputs and the predictions of the Driverless AI model. Then the parameters of this local GLM can be used to generate approximate, local explanations of the Driverless AI model.

mli_use_raw_features

Use Original Features for Surrogate Models :open:

Specify whether to use original features or transformed features in the surrogate model for the new interpretation. This is enabled by default.

Note: When this setting is disabled, the K-LIME clustering column and quantile binning options are unavailable.

mli_vars_to_pdp

Number of Features for Partial Dependence Plot

Specify the maximum number of features to use when building the Partial Dependence Plot. Use -1 to calculate Partial Dependence Plot for all features. By default, this value is set to 10.

mli_nfolds

Cross-validation Folds for Surrogate Models

Specify the number of surrogate cross-validation folds to use (from 0 to 10). When running experiments, Driverless AI automatically splits the training data and uses the validation data to determine the performance of the model parameter tuning and feature engineering steps. For a new interpretation, Driverless AI uses 3 cross-validation folds by default for the interpretation.

mli_qbin_count

Number of Columns to Bin for Surrogate Models

Specify the number of columns to bin for surrogate models. This value defaults to 0.

mli_sample_size

Sample Size for Surrogate Models

When the number of rows is above this limit, sample for surrogate models. The default value is 100000.

mli_num_quantiles

Number of Bins for Quantile Binning

Specify the number of bins for quantile binning. By default, this value is set to -10.

mli_dia_sample_size

Sample Size for Disparate Impact Analysis

When the number of rows is above this limit, sample for Disparate Impact Analysis (DIA). The default value is 100000.

mli_pd_sample_size

Sample Size for Partial Dependence Plot

When number of rows is above this limit, sample for the Driverless AI partial dependence plot. The default value is 25000.

mli_pd_numcat_num_chart

Unique Feature Values Count Driven Partial Dependence Plot Binning and Chart Selection

Specify whether to use dynamic switching between PDP numeric and categorical binning and UI chart selection in cases where features were used both as numeric and categorical by the experiment. This is enabled by default.

mli_pd_numcat_threshold

Threshold for PD/ICE Binning and Chart Selection

If mli_pd_numcat_num_chart is enabled, and if the number of unique feature values is greater than the threshold, then numeric binning and chart is used. Otherwise, categorical binning and chart is used. The default threshold value is 11.

mli_sa_sampling_limit

Sample Size for Sensitivity Analysis (SA)

When the number of rows is above this limit, sample for Sensitivity Analysis (SA). The default value is 500000.

klime_cluster_col

k-LIME Clustering Columns

For k-LIME interpretations, optionally specify which columns to have k-LIME clustering applied to.

Note: This setting is not found in the config.toml file.

qbin_cols

Quantile Binning Columns

For k-LIME interpretations, specify one or more columns to generate decile bins (uniform distribution) to help with MLI accuracy. Columns selected are added to top n columns for quantile binning selection. If a column is not numeric or not in the dataset (transformed features), then the column will be skipped.

Note: This setting is not found in the config.toml file.