Model Interpretation on Spark

Interpretable Machine Learning

Interpretable Machine Learning helps developers, data scientists and business stakeholders in the organization gain a comprehensive understanding of their machine learning models. It can also be used to debug models, explain predictions and enable auditing to meet compliance with regulatory requirements.

Why run model interpretation on Spark

Model-agnostic interpretation methods can be computationally expensive due to the multiple evaluations needed to compute the explanations. Model interpretation on Spark enables users to interpret a black-box model at massive scales with the Apache Spark™ distributed computing ecosystem. Various components support local interpretation for tabular, vector, image and text classification models, with two popular model-agnostic interpretation methods: LIME and Kernel SHAP.

Usage

Both LIME and Kernel SHAP are local interpretation methods. Local interpretation explains why does the model predict certain outcome for a given observation.

Both explainers extends from org.apache.spark.ml.Transformer. After setting up the explainer parameters, simply call the transform function on a DataFrame of observations to interpret the model behavior on these observations.

To see examples of model interpretability on Spark in action, take a look at these sample notebooks:

• Tabular Explainers
• Image Explainers
• Text Explainers

	Tabular models	Vector models	Image models	Text models
LIME explainers	TabularLIME	VectorLIME	ImageLIME	TextLIME
Kernel SHAP explainers	TabularSHAP	VectorSHAP	ImageSHAP	TextSHAP

Common local explainer params

All local explainers support the following params:

Param	Type	Default	Description
targetCol	String	"probability"	The column name of the prediction target to explain (i.e. the response variable). This is usually set to "prediction" for regression models and "probability" for probabilistic classification models.
targetClasses	Array[Int]	empty array	The indices of the classes for multinomial classification models.
targetClassesCol	String		The name of the column that specifies the indices of the classes for multinomial classification models.
outputCol	String		The name of the output column for interpretation results.
model	Transformer		The model to be explained.

Common LIME explainer params

All LIME based explainers (TabularLIME, VectorLIME, ImageLIME, TextLIME) support the following params:

Param	Type	Default	Description
regularization	Double	0	Regularization param for the underlying lasso regression.
kernelWidth	Double	sqrt(number of features) * 0.75	Kernel width for the exponential kernel.
numSamples	Int	1000	Number of samples to generate.
metricsCol	String	"r2"	Column name for fitting metrics.

Common SHAP explainer params

All Kernel SHAP based explainers (TabularSHAP, VectorSHAP, ImageSHAP, TextSHAP) support the following params:

Param	Type	Default	Description
infWeight	Double	1E8	The double value to represent infinite weight.
numSamples	Int	2 * (number of features) + 2048	Number of samples to generate.
metricsCol	String	"r2"	Column name for fitting metrics.

Tabular model explainer params

All tabular model explainers (TabularLIME, TabularSHAP) support the following params:

Param	Type	Default	Description
inputCols	Array[String]		The names of input columns to the black-box model.
backgroundData	DataFrame		A dataframe containing background data. It must contain all the input columns needed by the black-box model.

Vector model explainer params

All vector model explainers (VectorLIME, VectorSHAP) support the following params:

Param	Type	Default	Description
inputCol	String		The names of input vector column to the black-box model.
backgroundData	DataFrame		A dataframe containing background data. It must contain the input vector column needed by the black-box model.

Image model explainer params

All image model explainers (ImageLIME, ImageSHAP) support the following params:

Param	Type	Default	Description
inputCol	String		The names of input image column to the black-box model.
cellSize	Double	16	Number that controls the size of the super-pixels.
modifier	Double	130	Controls the trade-off spatial and color distance of super-pixels.
superpixelCol	String	"superpixels"	The column holding the super-pixel decompositions.

Text model explainer params

All text model explainers (TextLIME, TextSHAP) support the following params:

Param	Type	Default	Description
inputCol	String		The names of input text column to the black-box model.
tokensCol	String	"tokens"	The column holding the text tokens.

TabularLIME

Param	Type	Default	Description
categoricalFeatures	Array[String]	empty array	The name of columns that should be treated as categorical variables.

For categorical features, TabularLIME creates new samples by drawing samples based on the value distribution from the background dataset. For numerical features, it creates new samples by drawing from a normal distribution with mean taken from the target value to be explained, and standard deviation taken from the background dataset.

TabularSHAP

No additional params are supported.

VectorLIME

No additional params are supported.

VectorLIME assumes all features are numerical, and categorical features are not supported in VectorLIME.

VectorSHAP

No additional params are supported.

ImageLIME

Param	Type	Default	Description
samplingFraction	Double	0.7	The fraction of super-pixels to keep on during sampling.

ImageLIME creates new samples by randomly turning super-pixels on or off with probability of keeping on set to SamplingFraction.

ImageSHAP

No additional params are supported.

TextLIME

Param	Type	Default	Description
samplingFraction	Double	0.7	The fraction of word tokens to keep on during sampling.

TextLIME creates new samples by randomly turning word tokens on or off with probability of keeping on set to SamplingFraction.

TextSHAP

No additional params are supported.

Result interpretation

LIME explainers

LIME explainers return an array of vectors, and each vector maps to a class being explained. Each component of the vector is the coefficient for the corresponding feature, super-pixel, or word token from the local surrogate model.

• For categorical variables, super-pixels, or word tokens, the coefficient shows the average change in model outcome if this feature is unknown to the model, if the super-pixel is replaced with background color (black), or if the word token is replaced with empty string.
• For numeric variables, the coefficient shows the change in model outcome if the feature value is incremented by 1 unit.

SHAP explainers

SHAP explainers return an array of vectors, and each vector maps to a class being explained. Each vector starts with the base value, and each following component of the vector is the Shapley value for each feature, super-pixel, or token.

The base value and Shapley values are additive, and they should add up to the model output for the target observation.

Base value

• For tabular and vector models, the base value represents the mean outcome of the model for the background dataset.
• For image models, the base value represents the model outcome for a background (all black) image.
• For text models, the base value represents the model outcome for an empty string.