OneVsRestClassification with GridSearchCV in Sklearn

Question

When I tried to search the optimal C and gamma in rbf kernel SVM by:

params = dict(C = C_range, gamma = gamma_range)
clf = GridSearchCV(OneVsRestClassifier(SVC()),params, cv = 5)

It returns the error says C is not the parameter of OneVsRestClassifier. What is the proper way to achieve the grid search on the parameters with multiclass SVM then?

Answer 1

Grid search uses a custom nested attribute syntax for this:

params = dict(estimator__C=C_range, estimator__gamma=gamma_range)

The name estimator corresponds to the OneVsRestClassifier constructor parameter. Note the double underscores.