-
Notifications
You must be signed in to change notification settings - Fork 254
Open
Description
Hi - thanks for the great library.
E-MNIST can be split into various kinds, depending on whether class-balance is needed or not.
The balanced dataset header in this link shows that this is derived from the ByMerge split, and finally has 47 classes, unlike 62 (unbalanced).
Does each user in the F-EMNIST dataset have balance across classes? Directly by using your code, I got an unbalanced dataset per user with many classes having as low as 1 image within that task.
Is there a way to generate class-balanced user splits from the code?
Thank you
Metadata
Metadata
Assignees
Labels
No labels