Using built-in evaluation in rllib


I am using the built-in evaluation functionality in ray rllib by setting evaluation_interval in the trainer configuration to a number.
is there a keyword to request an additional evaluation on the final iteration (independent of the choice of evaluation_interval)?

Thank you in advance!

Hi @dbk80 ,

I don’t know of any configuration setting that ensures you an evaluation at the end of training.

What you can always do though is to call trainer.evaluate() and evaluate “manually”.

1 Like