The correct answer is: C. are useful mainly when the learning machines are “black boxes”.
Wrapper methods are a type of hyper-parameter selection method that uses a model to evaluate the performance of different sets of hyperparameters. This means that they can be used to select hyperparameters for any type of learning machine, even if the learning machine is a “black box” (i.e., if the inner workings of the learning machine are not known).
Wrapper methods are computationally expensive, so they should only be used when necessary. However, they can be very effective in finding the best set of hyperparameters for a given learning machine.
Option A is incorrect because wrapper methods are not always computationally efficient. They can be very expensive to run, especially if the learning machine is complex.
Option B is incorrect because wrapper methods can be used even if there are other options available. In fact, they can be more effective than other methods in some cases.
Option D is incorrect because wrapper methods can be very useful in some cases. They should not be avoided altogether.