Deploying logistic regression model with predetermined coefficients in R

Question

I have a logistic regression model with the coefficients already determined and I want to deploy in R.

I know it is extremely simple to just write my own function to do it, but I'm curious if there is some existing functionality that I am missing that's even simpler?

Basically I am looking to use something like the predict() functionality of glm with my own coefficients rather than fitting the model in R.

Answer 1

You can use the matrix form to do this:

logitp_est <- sum(c(1, values) * coefficients)

If you want the probability,

prob_est <- 1 / (1 + exp(-1 * logitp_est))

or the built in plogis():

prob_est <- plogis(logitp_est)

If you want the classification:

class_est <- logitp_est > 0

Answer 2

It's not recommended, but you could always change the coefficients by hand.

iris2 <- iris[, 1:4]; iris2$dep_var <- as.integer(ifelse(iris2$Sepal.Length > 5, 1, 0))
x <- glm(dep_var ~ ., family = binomial(link = logit), iris2)
x$coefficients
# (Intercept)  Sepal.Length   Sepal.Width  Petal.Length   Petal.Width
# -1990.9311682   392.5953392     2.0776581     0.5389770     0.9594286
predict(x, iris2[1, ])
#          1
#   19.52332
x$coefficients['Sepal.Length'] <- 393
predict(x, iris2[1, ])
#        1
# 21.58709

Note this will likely screw with things like summary(x).