Open HWRix opened 9 years ago
I don't agree with the terminology: In the leave-one-out test the left-out object is the "validation" object, not the "test" object. In the odd terminology of statistics, all of the survey objects are "test" objects. It is bad terminology but it is the reality. We can adopt or ignore, but we definitely should not use "test" for the validation objects.
ISSUE -- I would like to triple-check that we are on the same page regarding terminology:
We have REFERENCE objects, with trusted REFERENCE labels. Subsets of these reference objects are used in a) the TRAINING STEP b) the TEST STEP (leave-one-out test) The objects to which the test step is applied are the SURVEY OBJECTS
The whole process is called LABEL TRANSFER (when do we use that). I have no strong feelings about the nomenclature, just sting feelings about consistent use.
On that note: we never use anything but {\it The Cannon}, that is the term for the method (even if it catches on, others will drop the The)