get_influence: Given a 'model' object, rank the input predictors (and...

In BaTFLED3D: Bayesian Tensor Factorization Linked to External Data

Description

If method is 'add' then the baseline prediction is made using just the constant coefficients (if used) and the mean squared error (MSE) is measured between the baseline and predictions made with each predictor added alone (univariate analysis).

Usage

get_influence(m, d, method = "sub", interactions = TRUE)

Arguments

m	Tucker_model or CP_model object
d	input_data object
method	string 'sub' or 'add' indicating whether to start with a full or empty feature vector and remove or add features to judge their influence.
interactions	logical indicating whether to get influence for two-way interactions between predictors (def: sub)

Details

If method is 'sub' then the baseline is made using all predictors and MSE measured for predictions made with each predictor removed.

If interactions==TRUE then MSE for predictions made with predictors for each mode interacting are measured

BaTFLED3D documentation built on May 2, 2019, 2:38 p.m.