Grid search and cross validation

Example of grid search with an elastic net model

Import the model for which parameters are to be optimized and create an instance.

from sklearn.linear_model import ElasticNet
base_elastic_net_model = ElasticNet()

Create a dictionary with the parameters to test. The strings must match hyperparams in the model being optimized.

param_grid = {'alpha':[0.1,1,5,10,50,100],'l1_ratio':[.1,.5,.7,.95,.99,1]}

Pass all this into a GridSearchCV object, and then treat the object like the model itself:

from sklearn.model_selection import GridSearchCV
grid_model = GridSearchCV(estimator=base_elastic_net_model,
                         param_grid=param_grid,
                         scoring='neg_mean_squared_error',
                         cv=5,verbose=2)

grid_model.fit(X_train,y_train)

We can then check which of the provided parameters performed better in the cross validation.

grid_model.best_estimator_
grid_model.best_params_

Checking the results

pd.DataFrame(grid_model.cv_results_)

Cross_val_score

from sklearn.model_selection import cross_val_score
scores = cross_val_score(model,X_train,y_train,scoring='neg_mean_squared_error',cv=5)
abs(scores.mean())

cross_validate

from sklearn.model_selection import cross_validate
scores = cross_validate(model,X_train,y_train,scoring=['neg_mean_squared_error',
                                                      'neg_mean_absolute_error'],cv=10)