1
$\begingroup$

I read this link Bergstra-Bengio in which it is discussed why 60 parameters randomly chosen are enough.

But I do not fully understand the formula used

$$1−(1−0.05)n>0.95$$

$$n⩾60$$

I have read the paper of Bergstra and Bengio; but I have not found anything about that formula. Also, I was wondering if RandomSearch is only good for Neural Networks; as I am using RandomForest, XGBoost and SVM.

Thank you

$\endgroup$
4
  • $\begingroup$ Did you read the explanation at this answer? What part of it do you not understand? stats.stackexchange.com/questions/160479/… $\endgroup$
    – Sycorax
    Feb 24 '17 at 0:49
  • $\begingroup$ @Sycorax Hi. Yes I did read the answer, I also read Bengio's paper. I simply do not understand the formula. Can the 60 observations be used with any number of parameters? $\endgroup$
    – Aizzaac
    Feb 24 '17 at 17:09
  • 2
    $\begingroup$ Don't you mean "$1 - (1-0.05)^n$"? $\endgroup$
    – whuber
    Feb 24 '17 at 17:57
  • $\begingroup$ @Sycorax Is it still true if I optimize 2 or 8 parameters? $\endgroup$
    – Aizzaac
    Feb 24 '17 at 19:25