CODEWITHSUNDEEP
pythonscikit-learn
Thomas Johnson
Thomas Johnson

Reputation: 11718

How to re-fit best classifier in sklearn

Using scikit-learn, I fit a classifier using Grid Search like this:

from sklearn.svm import SVC
param_grid = { 
    'C': [1e-2, 0.1, 1.0],
    'gamma': [1e-4, 1e-3, 1e-2],
    'class_weight': ['auto']
}

clf = SVC()
gs = grid_search.GridSearchCV(clf, param_grid, cv=3, n_jobs=12)
gs.fit(x_train, y_train)

I now want to re-train the classifier using the best parameters found and the extra argument probability=True. How can I re-fit the classifier using the best parameters, plus the extra parameter probability?

Upvotes: 4

Views: 3369

Answers (3)

eickenberg
eickenberg

Reputation: 14377

Try

best_estimator = grid_search.best_estimator_.set_params(probability=True)

You can also clone it to be sure no other part of your code reuses this estimator.

Upvotes: 0

ely
ely

Reputation: 77484

You can also use the set_params method for an instance of SVC and modify the probability attribute before calling fit.

from sklearn import svm, grid_search

x_train = np.random.randn(10,5)
y_train = np.random.randint(0, 2, size=(10,1))

param_grid = { 
    'C': [1e-2, 0.1, 1.0],
    'gamma': [1e-4, 1e-3, 1e-2],
    'class_weight': ['auto']
}

svc1 = svm.SVC()
gs = grid_search.GridSearchCV(svc1, param_grid, cv=3, n_jobs=12)
gs_fitted = gs.fit(x_train, y_train)

svc2 = svm.SVC(probability=True)
# or manually set svc2.probability = True before ever calling svc2.fit

svc2.set_params(**gs_fitted.best_params_)
svc2.fit(x_train, y_train)

Upvotes: 3

Matthias Ossadnik
Matthias Ossadnik

Reputation: 911

You can use gs.best_params_ to get the parameters, then create a new classifier like this

clf = SVC(probability=True, **gs.best_params_)

Upvotes: 5

Related Questions