It would be good if the Trial object had a user_suggested field which would allow us to track which trials are coming from user suggested parameter values.
This would help the user themselves in the case of non-in-order executions.
This would also allow the optimiser to have the ability not to drop them from distribution estimation even if linear_forgetting would otherwise make them remain unused.
Describe the solution you'd like
It would be good if the Trial object had a
user_suggestedfield which would allow us to track which trials are coming from user suggested parameter values.This would help the user themselves in the case of non-in-order executions. This would also allow the optimiser to have the ability not to drop them from distribution estimation even if linear_forgetting would otherwise make them remain unused.