Select-regress-and-round

srr-example.R

# Assuming a (logistic) model with k features fit in R with the variable name "model", 
# generate a simple rule that uses integers in the range [-M, M]: 
# (note that M is set to 3 in this example)
M <- 3
model_coefs <- coef(model)
scaled_coefs <- (model_coefs / max(model_coefs)) * M
rounded_coefs <- round(scaled_coefs)

# The k features to be included in the initial model is best determined by domain expertise, 
# i.e., what are the minimal features that expert(s) in the field believes to be predictive 
# of whatever you are trying to predict? 
# Although, in the absence of domain expertise, you could use step-wise variable selection methods 
# e.g., leaps