如何将现有系数加载到sklearn SVM分类器？

Question

I trained a SVM classifier using sklearn.svm.SVC , and stored the weights (coefficients). Then I loaded them and tried to inject them into a new instance of sklearn.svm.SVC , but could not do it because it seems that the attribute coef_ is read only..

from sklearn import svm
import pickle

modelSVM = svm.SVC(kernel='linear')
weights = pickle.load(open(weights_path, 'rb'))

modelSVM.coef_ = weights

I expect to have a model with the weights I loaded as new coefficients, but I get this message:

AttributeError: 'SVC' object has no attribute 'dual_coef_'

Which might be due to the fact that coef_ is not the only field needed by the classifier. So I tried to train and then clone the classifier before injecting weights:

modelSVM.fit(X, labels)
modelSVM = clone(modelSVM)
modelSVM.coef_ = weights

It gives the output:

"Exception has occurred: AttributeError
can't set attribute"

Answer 1

Instead of loading the weights, which is not possible as pointed out by @Mechanic in the comments. You can try to save the model and load the model back. Please refer to this link .