[Python] Change in random state value changes the model accuracy

While testing my linear regression model, I find that changing the random_state parameter in train_test_split changes the model accuracy.

To elaborate, my train test split is as follows:

boston_data = load_boston()
X = pd.DataFrame(boston_data.data, columns=boston_data.feature_names)
y = pd.DataFrame(boston_data.target, columns=['MEDV'])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)


And my model is as follows:

steps = [
('regr', Lasso())
]
pipeline = Pipeline(steps)

parameters = {
'regr__alpha': np.logspace(-4, -0.5, 40)
}

grid = GridSearchCV(pipeline, param_grid=parameters, cv=10, n_jobs=-1)
grid.fit(X_train, y_train)
grid.score(X_test, y_test)


Now, for example, for the current scenario (where random_state = 42) the test score is 0.725. However, if I change it to 43, the test score comes down to 0.7.

I understand that random_state signifies what set of data will be in test and training set. Having said that, I want to know, if there's way to get a stable result?

Thanks!

Continue reading...