Evaluation taking much longer than the running epochs about deep-symbolic-optimization HOT 4 CLOSED

I did some further debugging on this adding print statements to see at what point the evaluation halts (see image below) and it appears to have some sort of a problem at the results_add line below:
results_add = logger.save_results(positional_entropy, top_samples_per_batch, r_history, pool, epoch, nevals)

from deep-symbolic-optimization.

See the template config for the control task.

During training, each sampled expression is used as a policy in the environment for n_episodes_train episodes.

During eval, each tested expression is used for n_episodes_test episodes, which is 100x longer by default, since looks like you didn't change this parameter in your config.

For debugging purposes, I'd just set both n_episodes_train and n_episodes_test to 1.

from deep-symbolic-optimization.

Regarding the prints, you might want to add flush=True to be certain that it's not passing more of your DEBUG points. Or use an actual breakpoint or something more robust than printouts.

from deep-symbolic-optimization.

Ah I see.
You are right I missed out on that within my config.
It does seem to reach further now that I have set it to 1 for both.
Thanks.

from deep-symbolic-optimization.