log_experiment should be able to run and log hypotheses without logging learner state and without specifying a value for observer_path. However, what actually happens is that it crashes (with a NameError, I think) whenever it tries to do the hypothesis logging, because log_learner_state isn't getting checked at the point where we use observer_path.
This is not too terrible a problem. Leaving log_learner_state on when it's not needed doesn't hurt much -- it only costs a small amount of disk space and write time.
log_experiment
should be able to run and log hypotheses without logging learner state and without specifying a value forobserver_path
. However, what actually happens is that it crashes (with aNameError
, I think) whenever it tries to do the hypothesis logging, becauselog_learner_state
isn't getting checked at the point where we useobserver_path
.This is not too terrible a problem. Leaving
log_learner_state
on when it's not needed doesn't hurt much -- it only costs a small amount of disk space and write time.This looks like an easy fix.