Generate Interpretation

Synopsis

This operator allows you to use interpretation algorithm to explain your predictions

Description

There are several algorithms around which help you to understand which properties of a given example was most 'impactful' for the model to decide how it decided. This operator allows you to use Explain Predictions, LIME (kernel)SHAP and Shapley. For more information about the algorithms we recommend Interpretable Machine Learning by Christoph Molnar (https://christophm.github.io/interpretable-ml-book/).

Input

mod

The model whose predictions are to be explained.

training

The training data.

testing

The testing data is the data you want to get an interpretation for.

Output

example set

The input data set with an additional column 'interpretation' giving you the interpretation.

importance

A data set with the detailed interpretations for each row.

global weights

The global weights of the model. This is the average of the absolute local weights.

mod

The original model passed through.

Parameters

Algorithm

The algorithm which is used to interpret your prediction

Sample size

Most algorithm use some sample size generation internally. For LIME and Explain Predictions this is the number of random sample drawn. In case of Shapely and kernelSHAP this is the number of permutations.

Redraw local samples

Only available for LIME. If checked the operator will draw new random samples in each iteration. If deactivated a set of examples is generated once and used for the interpretation of each example.

Explanation algorithm

Only available LIME. Defines what algorithm is used on the local neighbourhood to generate your interpretation. The original paper uses Linear Regression models. Explain Prediction uses (besides some other adoption) a correlation measure to determine the interpretation.

Maximal explaining attributes

Defines how many attributes are shown in the interpretation column of the result. The full information is always shown at the importance port.