can the training log/json update more frequently than every 2k iters to give the user more feedback. maybe configureable?