linpred {sqlscore}

Unevaluated prediction expressions for models

Description

Generate an unevaluated call corresponding to the predict step of the passed model. The call represents the linear predictor in terms of elementary functions on the underlying column names. Before translation into SQL, it should have a response function applied by score_expression (which may be a no-op in the case of the identity response).

Usage

linpred(mod)

Arguments

mod

A supported model object.

Value

An unevaluated R call object representing the linear predictor.

Warning

The Binomial models in glmboost return coefficients which are 1/2 the coefficients fit by a call to glm(..., family=binomial(...)), because the response variable is internally recoded to -1 and +1. sqlscore multiplies the returned coefficients by 2 to put them back on the same scale as glm, and adds the glmboost offset to the intercept before multiplying.

Examples

# A Gaussian GLM including factors
mod <- glm(Sepal.Length ~ Sepal.Width + Petal.Length + Petal.Width + Species,
           data=datasets::iris)
linpred(mod)

# A binomial GLM - linear predictor is unaffected
mod <- glm(Sepal.Length > 5.0 ~ Sepal.Width + Petal.Length + Petal.Width + Species,
           data=datasets::iris, family=binomial("logit"))
linpred(mod)

#With formula operators
x <- matrix(rnorm(100*20),100,20)
colnames(x) <- sapply(1:20, function(x) paste0("X", as.character(x)))
x <- as.data.frame(x)
mod <- glm(X2 ~ X3 + X5 + X15*X8, data=x)
linpred(mod)