model_support: Methods for extending anchors model support

In viadee/anchorsOnR: AnchorsOnR: High-Precision Model-Agnostic Explanations in R

Description

In order to have 'anchors' support for your model of choice 'anchors' needs to be able to get predictions from the model in a standardised way, and it needs to be able to know whether it is a classification or regression model. For the former it calls the 'predict_model()' generic which the user is free to supply methods for without overriding the standard 'predict()' method. For the latter the model must respond to the 'model_type()' generic.

Usage

predict_model(x, newdata, type, ...)

model_type(x, ...)

Arguments

x	A model object
newdata	The new observations to predict
type	Either ''raw'‘ to indicate predicted values, or '’prob'' to indicate class probabilities
...	passed on to 'predict' method

Value

A data.frame in the case of ‘predict_model()'. If 'type = ’raw'' it will contain one column named ''Response'' holding the predicted values. If ‘type = ’prob'' it will contain a column for each of the possible classes named after the class, each column holding the probability score for class membership. For ‘model_type()' a character string. Either '’regression'' or ''classification'' is currently supported.

Supported Models

Out of the box, 'anchors' supports the following model objects:

- 'WrappedModel' from mlr - 'H2OModel' from h2o - 'lda' from MASS (used for low-dependency examples)

If your model is not one of the above you'll need to implement support yourself. For that you'll need to implement a 'predict_model()' method and potentially a 'model_type()' method (if the latter is omitted the model should be wrapped in [as_classifier()]/[as_regressor()], everytime it is used in [anchors()]).

viadee/anchorsOnR documentation built on Nov. 22, 2019, 5:24 p.m.