Skip to content

Instantly share code, notes, and snippets.

@jongbinjung
Created May 16, 2017 18:20
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save jongbinjung/34e225663b453c6bec4ea509f9baf2a5 to your computer and use it in GitHub Desktop.
Save jongbinjung/34e225663b453c6bec4ea509f9baf2a5 to your computer and use it in GitHub Desktop.
Select-regress-and-round
# 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
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment