The idea behind RandomizedSearchCV is that it works the same as GridSearchCV, but instead of trying all the combinations, it tries a randomly sampled subset.

For example, this param_grid will have 100 combinations:

param_grid = {'n_neighbors': [1, 3, 5, 7, 9, 12, 15, 17, 20, 25],
                         'weights': ['distance', 'uniform'],
                         'p': [1, 2, 3, 4, 5]}

The GridSearchCV would try all of them, which is time-consuming. With RandomizedSearchCV, you can try only a randomly chosen subset of, say, 20 combinations. It usually leads to a little worse result, but works much faster.

You can control the number of combinations to be tested using the n_iter argument (set to 10 by default). Apart from that, working with it is the same as with GridSearchCV.