0

I have to solve a simple binary classification problem using H2O AutoML. I'd like to know if the parameters sort_metric and stopping_metric can somehow influence the order of the trained model.

I try to change these two parameters using both AUC or AUCPR, but the performances are almost identical.

My principal objective is to obtain the best algorithms in terms of AUCPR, so I would like to somehow influence the order of the trained models.

Does someone know how can I do so?

A1010
  • 360
  • 5
  • 18
  • By "order" do you mean to want to change the relative ranking of models in the leaderboard? If so `sort_metric` is exactly what you want: http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/sort_metric.html But that won't change the accuracy/predictions of the individual models at all. – Darren Cook Jun 24 '20 at 08:13
  • Yes, exactly. I understand that ```sort_metric``` is the right parameter in order to change leaderboard order, but it doesn't affect directly the training. Anyway, I think it would be more effective a change directly into the loss function. Is it possible to specify such a parameter in AutoML? – A1010 Jun 24 '20 at 12:51

0 Answers0