RandomizedSearch vs GridSearch


If the region of good hyperparameters is what I have circled in red, the randomized-search would give us a worth result than the grid-search.

That’s why I still don’t understand why it is good to avoid the regularity imposed by the grid. Could you be more precise please ?

Well, I think the orange points in this picture is about to the final result of hyperparameters searched by “randomized-search”. These points reflects the candidates that was searched based on the criteria defined in “randomized-search” and hence the “red circle” may not attain important information and has been avoided by “randomized-search”.

@saeedsltm is right. I put another figure [1] where it might be better explained:

There you want to get the best results (top of the green bell) and you see that you don’t select it with a grid-like approach while a randomized search would be more likely to explore it.

[1] taken from Grid Search vs Random Search. In this article, we will focus on two… | by Deepak Senapati | Medium

1 Like