Scikit-Learn Configure RandomizedSearchCV "random_state" Parameter

The random_state parameter in RandomizedSearchCV controls the randomness of the hyperparameter search process. It ensures reproducibility of the search results by fixing the random seed used for generating parameter combinations.

The default value for random_state is None, which means the search process will be different each time the code is run. Setting random_state to an integer value will make the search deterministic, producing the same results across multiple runs with the same random seed.

As a heuristic, set random_state to a fixed integer when you need reproducible results, such as for debugging or comparison purposes. Use None or different integer values to explore different random subsets of the hyperparameter space.

from sklearn.datasets import make_classification
from sklearn.svm import SVC
from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import randint

# Generate a synthetic binary classification dataset
X, y = make_classification(n_samples=100, n_features=10, n_informative=5, n_redundant=5, random_state=42)

# Define a parameter distribution for SVC hyperparameters
param_dist = {'C': randint(1, 10), 'kernel': ['linear', 'rbf']}

# Create a base SVC model
svc = SVC()

# List of random_state values to test
random_states = [None, 42, 123]

for rs in random_states:
    # Run RandomizedSearchCV with the current random_state value
    search = RandomizedSearchCV(svc, param_dist, n_iter=5, cv=3, random_state=rs)
    search.fit(X, y)

    print(f"Best score for random_state={rs}: {search.best_score_:.3f}")
    print(f"Best parameters for random_state={rs}: {search.best_params_}")
    print()

Running the example gives an output like:

Best score for random_state=None: 0.880
Best parameters for random_state=None: {'C': 6, 'kernel': 'rbf'}

Best score for random_state=42: 0.890
Best parameters for random_state=42: {'C': 7, 'kernel': 'rbf'}

Best score for random_state=123: 0.890
Best parameters for random_state=123: {'C': 7, 'kernel': 'rbf'}

The steps are as follows:

Generate a synthetic binary classification dataset using make_classification().
Define a parameter distribution dictionary param_dist for SVC hyperparameters.
Create a base SVC model.
Iterate over different random_state values (None, 42, 123).
For each random_state value:
- Run RandomizedSearchCV with 5 iterations and 3-fold cross-validation.
- Print the best score and best parameters for the current random_state value.

See Also