Open EricR86 opened 7 years ago
Original comment by Michael Hoffman (Bitbucket: hoffman, GitHub: michaelmhoffman).
Probably should be called likelihood
.
Original comment by Rachel Chan (Bitbucket: rcwchan).
Should minibatch automatically test likelihood on a validation set? If yes, should the region be automatically inferred or should it be required to be specified using something such as --validation-coords
?
If automatically inferred, what size should the validation set be? I know we previously said '10x the training set', but what if the minibatch fraction specified is extremely large (ie, 0.10 or greater)?
Original comment by Michael Hoffman (Bitbucket: hoffman, GitHub: michaelmhoffman).
How validation works for minibatch is a separate issue than segway likelihood
which will essentially be used for test. Please open another enhancement request or discuss on segway-internal
.
Original report (BitBucket issue) by Rachel Chan (Bitbucket: rcwchan).
A new verb (ie, 'validate') could be added such that Segway tests likelihood on
--include-coords
; that is, run gmtKJTwith probE on the specified region(s).