Suppose that 2 different trials are stopped, one of them has a terribly bad objective score, the other one has a decent objective score (just a bit worse than the stopping threshold). Does Sherpa take those scores into account when sampling new parameters? In other words, does it use the information obtained frome pruned (i.e., early stopped) trials in the optimization process or is that information discarded/ignored? I ask because considering that information may avoid sampling (i.e., suggesting) unpromising hyperparameters (that were pruned) multiple times.
Suppose that 2 different trials are stopped, one of them has a terribly bad objective score, the other one has a decent objective score (just a bit worse than the stopping threshold). Does Sherpa take those scores into account when sampling new parameters? In other words, does it use the information obtained frome pruned (i.e., early stopped) trials in the optimization process or is that information discarded/ignored? I ask because considering that information may avoid sampling (i.e., suggesting) unpromising hyperparameters (that were pruned) multiple times.