RandomizedSearchCV

RandomizedSearchCV is a hyperparameter optimization tool in scikit-learn that performs randomized search with cross-validation. It aims to identify a good combination of hyperparameters for an estimator by evaluating a fixed number of parameter settings sampled from specified distributions or sequences. The search uses cross-validated performance to assess each candidate configuration and selects the one that yields the best score according to the configured scoring metric. After fitting, the best_estimator_ can be refit on the full training data if requested.

Key features include the param_distributions argument, which describes the parameter space as distributions or lists of values; n_iter, which controls how many configurations are sampled; and standard options such as cv, scoring, refit, random_state, n_jobs, and verbose. Parameter distributions may use distributions from scipy.stats or explicit lists, enabling both continuous and discrete searches. Random state ensures reproducibility of the sampling.

Compared with GridSearchCV, RandomizedSearchCV does not exhaustively try all combinations but instead samples a fixed number of configurations. This makes it more scalable for large or high-dimensional hyperparameter spaces and often faster while still delivering strong performance. However, there is a risk of missing the optimal configuration, and results can vary between runs if a random state is not fixed.

Typical outputs include best_params_, best_score_, and cv_results_. The estimator with the best settings can be retrieved via best_estimator_, and, if refit is enabled, this estimator is trained on the entire training dataset. This approach is widely used when the hyperparameter space is large or expensive to evaluate.