I want to use Grid Search for finding optimal hyperpameters for Random Forest. My questions are:

  1. Which Random Forest hyperparameters are considered important for tuning?
  2. Which initial Random Forest hyperparameters values for Grid Search do you recommend?
  • $\begingroup$ What's an "initial hyperparameter value for grid search"? Grid search varies the hyperparameter values under optimization as a part of the search. The sequence doesn't matter, because all values in the grid are tried. $\endgroup$ – Sycorax Feb 21 at 2:54
  • $\begingroup$ @Sycorax Thank you for asking. What I mean by "initial hyperparameters values" is like for example initial values for hyperparameter1 are recommended 1,2,3 like this: 'hyperparameter1':[1,2,3], values for hyperparameter2 are recomended 0.1, 0.01, 0.001 like this 'hyperparameter2': [0.1, 0.01, 0.001] and so on for each hyperparameter. $\endgroup$ – vasili111 Feb 21 at 3:11
  • $\begingroup$ Oh so you're asking what values you should use for the grid? There's no single, correct answer to this because different problems will have different optimal configurations (this is why hyper-parameter search is necessary). If you have no limits on your computational budget (hours, memory, CPUs etc)., then you can just make the grid as large as you please and wait. If you have tighter constraints, then you'll need a coarser grid. There are some recommendations in the duplicate questions, but they're not guaranteed to be "correct," because such a guarantee is impossible. $\endgroup$ – Sycorax Feb 21 at 3:13