Why is random state required for ridge lasso regression classifiers

Question

class sklearn.linear_model.Lasso(alpha=1.0, fit_intercept=True, normalize=False, precompute=False, copy_X=True, max_iter=1000, tol=0.0001, warm_start=False, positive=False, random_state=None, selection=’cyclic’)

class sklearn.linear_model.Ridge(alpha=1.0, fit_intercept=True, normalize=False, copy_X=True, max_iter=None, tol=0.001, solver=’auto’, random_state=None)

Answer 1

This is because the regression coefficients of each variable are fitted in the Lasso. This can be done in a 'cyclic' manner, or by picking random variables at each iteration. The first has the attribute selection ='cyclic,' while the second has the attribute selection ='random.' Random numbers are used in the latter.

It's essential for Ridge regression if you want to fit the model with stochastic gradient descent, which uses subsampling. To do so, declare solver ='sag' or solver ='saga' during your model's initialization.