0

I see in the documentation that 100 labels are allowed per dataset, but does that mean that I am restricted to only 100 labels in my model or can I re-train the same model with 100 new labels and a new dataset?

Ethan Lebioda
  • 3
  • 3

1 Answers1

0

I work on automl at Google. You are restricted to a max of 100 labels for the model. Currently, re-training the model with model data is not supported. You'll be basically adding your new data to the old data and train from scratch. If you add more labels the second round, then you'll have > 100 labels, and that's above the limit. In your usecase, do you basically keep having new labels as you have more data?

Mona Attariyan
  • 191
  • 4
  • Potentially there would be new labels as new data comes in, but I was just wondering if you could compound new labels onto the model if re-trained, but from your answer it seems like it is not possible. Thank you very much! – Ethan Lebioda Oct 27 '18 at 16:20
  • As in start from a model with X number of labels and train from there to X+Y labels, right? – Mona Attariyan Oct 31 '18 at 18:45