josherrickson / postr Goto Github PK

Should postr enable some sort of attach-style (or Stata post-estimation style) functionality? To enable something like this:

m <- glm(...)
pr_attach(m)
pr_classify(threshold = .2)
pr_roc()

• Should definitely only allow one model to be "attached" at a time. This gets around the common attach problem of repeated attaches.
• Should not extend attach itself.
• Are there any other attach-related concerns?
• Should a model name be required? E.g. should pr_attach(glm(...)) fail?
• Needs a pr_summary() or equivalent.
• pr_info() to gives details about "attached" model.
• Terminology options:
  • attach
  • active
  • fixate
  • current
  • function name should probably be a verb?

Generalize rvfplot to enable easy plotting of residuals against other functions of x.

Against x is easy, against generic f(x)?

How should pr_rvfplot work with glm models e.g. logistic regression?

1. refuse to run.
2. refuse to run, but be over-rideable.
3. run with warning.
4. run

• effect size
• covariance matrix extraction
• AIC/BIC/etc
• Hosmer-lemeshow test
• ICC
• Youden index

Most functions should allow an optional newdata argument. If it is not given, the data is assumed to be that upon which the model was fit.