Hi, Igniters Welcome to the discussion about labels handling during ML training.
The problem is that all algorithms of binary classification are ready to handle the datasets marked with 0/1 labels and predict 0/1 labels without especial mapping. Also the algorithms don't handle situation with unknown labels during the updating and testing phases Umbrella ticket is created here https://issues.apache.org/jira/browse/IGNITE-9587 Also, I'd invite you to discuss 1) list of trainers to upgrade with this feature 2) how to handle unknown labels during the prediction/test phase 3) how to handle unknown labels during the update model phase (new data is coming during the next training and the results of next training should be merged with the results of the previous training) 4) Where to store metadata about labels during training phases Glad to hear your ideas
