Extract the training objective value for each candidate tree in the comparison, where a lower value indicates a better solution