woodwork icon indicating copy to clipboard operation
woodwork copied to clipboard

[Feature Request] Smarter variable type inference

Open rohan-gt opened this issue 4 years ago • 2 comments

Is it possible to add smarter variable type inference to detect all the different variable types that Featuretools supports like PhoneNumber, ZipCode etc. using regex or other rules?

rohan-gt avatar Jul 10 '20 09:07 rohan-gt

this is something we've been thinking about. i think it breaks into 2 categories, both of which would be valuable

  1. validation - does the data match the variable type
  2. inference - can we automatically determine and parse the correct type without user input

i think validation might be a bit easier to start with and would give us a starting point to think about how to do inference

kmax12 avatar Jul 10 '20 17:07 kmax12

Hmm validation could be done based on a random sample or to be more thorough scan all values and using a confidence threshold although the latter might be computationally expensive

rohan-gt avatar Jul 13 '20 16:07 rohan-gt