Which hyperparameter tuning method randomly selects combinations of hyperparameters?

The method that randomly selects combinations of hyperparameters is indeed random search. This technique involves sampling from a defined range of hyperparameters, allowing it to explore different combinations without the need to systematically exhaust every possible configuration, as is the case with grid search or exhaustive search.

One of the key advantages of random search is that it can often find good hyperparameter sets more efficiently than grid search. While grid search tests all possible combinations in a defined grid, it might overlook some potentially good combinations between the points it tested. Random search, on the other hand, can generate a diverse set of combinations, which might lead to discovering optimal performance with fewer evaluations.

The method is particularly beneficial in high-dimensional spaces where the grid becomes exponentially large, making grid search impractical. By randomly exploring the hyperparameter space, random search can significantly reduce computation time and improve model tuning outcomes.