add_predictions: Add black box predictions to generated dataset

In MI2DataLab/live: Local Interpretable (Model-Agnostic) Visual Explanations

Description

Add black box predictions to generated dataset

Usage

add_predictions(to_explain, black_box_model, data = NULL,
  predict_fun = predict, hyperparams = list(), ...)

Arguments

to_explain	List return by sample_locally function.
black_box_model	String with mlr signature of a learner or a model with predict interface.
data	Original data frame used to generate new dataset. Need not be provided when a trained model is passed in black_box_model argument.
predict_fun	Either a "predict" function that returns a vector of the same type as response or custom function that takes a model as a first argument, and data used to calculate predictions as a second argument and returns a vector of the same type as respone. Will be used only if a model object was provided in the black_box argument.
hyperparams	Optional list of (hyper)parameters to be passed to mlr::makeLearner.
...	Additional parameters to be passed to predict function.

Value

list of class "live_explorer" consisting of

data	Dataset generated by sample_locally function with response variable.
target	Name of the response variable.
model	Black box model which is being explained.
explained_instance	Instance that is being explained.
sampling_method	Name of used sampling method
fixed_variables	Names of variables which were not sampled
sdevations	Standard deviations of numerical variables

Examples

## Not run: 
# Train a model inside add_predictions call.
local_exploration1 <- add_predictions(dataset_for_local_exploration,
                                      black_box_model = "regr.svm",
                                      data = wine)
# Pass trained model to the function.
svm_model <- svm(quality ~., data = wine)
local_exploration2 <- add_predictions(dataset_for_local_exploration,
                                      black_box_model = svm_model)

## End(Not run)

MI2DataLab/live documentation built on Aug. 21, 2019, 6:05 p.m.