numpy - Why would reversing the order on my training set in Scikit learn change the predictions? -
i have training set in sci-kit learn i'm using perceptron algorithm. flipped training set's order, see happened, , i'm surprised seeing different results.
do know why case?
the perceptron algorithm implemented in scikit-learn sensitive data ordering nature sequential algorithm n_iter hard stopping criterion. in particular recommended shuffle data ensure examples of different classes have equal likelihood occur @ beginning , end of learning. classical stopping criterion of perceptron algorithm (stability of last epoch) not implemented in scikit-learn. if increase n_iter see less different outcomes .
Comments
Post a Comment