Building a Sentiment Classifier using Scikit-Learn

4.hyper_parameter_tuning_1.py

from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import uniform

X_train = X_train_bigram_tf_idf


# Phase 1: loss, learning rate and initial learning rate

clf = SGDClassifier()

distributions = dict(
    loss=['hinge', 'log', 'modified_huber', 'squared_hinge', 'perceptron'],
    learning_rate=['optimal', 'invscaling', 'adaptive'],
    eta0=uniform(loc=1e-7, scale=1e-2)
)

random_search_cv = RandomizedSearchCV(
    estimator=clf,
    param_distributions=distributions,
    cv=5,
    n_iter=50
)
random_search_cv.fit(X_train, y_train)
print(f'Best params: {random_search_cv.best_params_}')
print(f'Best score: {random_search_cv.best_score_}')