Optimal parameter selection

[HL-97]

In the process of tuning parameters, I found that after training the same number of epochs (only tested 10 times), the golden setting returned by Bayesian optimization is not the one with the smallest loss or the largest hit value. What is the cause of this? Thanks for your answer！

[HL-97]

So should I choose the golden setting returned by Bayesian optimization, or manually select the set of parameters with the lowest loss and the largest hits value by myself ?