0
$\begingroup$

Is there an accepted method for separating out a validation set in python? In R I would use the sample function. I have 4000 training instances as json and I want to save out a validation set. Should I just randomly pick indices and separate those out? Also is 30% of the training set a good size for the validation set?

$\endgroup$
1
$\begingroup$

I'm not fluent in Python so I'll stay out of code advices but for the percentage part, a 30% is a fairly standard ratio, as Andrew Ng points out in page 2 of the CS229 material for Model Selection, in http://cs229.stanford.edu/notes/cs229-notes5.pdf.

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$
  1. I don't think you'll get much advice on python code. I haven't seen much on this site.
  2. Random sampling is almost always the correct approach. (If you have clustering or events recorded over time or at different locations - then other sampling techniques may have added benefit.
  3. The issue of split sample has been addressed here a few times. Sample entries:
    Splitting the dataset into Testing,Cross Validation and Training Set
    Logistic regression performs better on validation data

Generally
-70/30 split commonly done. More traditional than anything. (though would be great if someone had a reference for this)
-Often sample size is inadequate to justify 70/30 split (see logistic regression... above).

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.