Scorecard_fit: Fit a model to the current set of variable transforms

In Zelazny7/binnr: Scorecard modeling in R

Description

fits a regularized regressoion model using the glmnet package.

Arguments

name	brief model name as character
description	character description describing the model
overwrite	should model be overwriiten if it already exists?
newdata	data.frame of independent variables. Default is to use the binned data.
y	target to fit the data to. Default is the y variable used for discretization.
w	optional weight variable
nfolds	number of k-folds with which to select the optimal lambda value
upper.limits	maximum value of fitted coefficients
lower.limits	minimimum value of fitted coefficients
alpha	type of regularization. Default is alpha == 1 for LASSO regression. Alpha of 0 is Ridge regression while anythin in between is the elastic net mixture.
family	response variable distribution. Default is "binomial".
...	additional arguments passed on to cv.glmnet

Details

the fit function first calls predict and substitutes the weight-of-evidence for all predictor variables. It then passes this matrix on to cv.glmnet. The coefficients of a binner model fit are restricted to [0,3]. This ensures there are no sign flips in the model coefficients and that the relationships observed on margin are retained in the final model. bin wrapper function.

Zelazny7/binnr documentation built on May 10, 2019, 1:55 a.m.