Unfortunately, Optuna does not yet support pruning in multi-objective setups.
To save some computational budget, time_pruning is already implemented, which prunes trials based on the predicted training time - if it exceeds a threshold, the trial is pruned. The main motivation for this is that training time is directly related to inference time, hence very slow trials are likely to yield very slow surrogates.
To save additional computational budget, an early stopping criterion which would end model.fit(...) prematurely when progress is slow could be implemented. Note that early-stopped trials are not pruned, they will still count as a proper trial.
Unfortunately, Optuna does not yet support pruning in multi-objective setups.
To save some computational budget, time_pruning is already implemented, which prunes trials based on the predicted training time - if it exceeds a threshold, the trial is pruned. The main motivation for this is that training time is directly related to inference time, hence very slow trials are likely to yield very slow surrogates.
To save additional computational budget, an early stopping criterion which would end model.fit(...) prematurely when progress is slow could be implemented. Note that early-stopped trials are not pruned, they will still count as a proper trial.