[FIXED] Can we predict y when y value is not numeric?

Issue

I am using support vector regressor. I want to predict personality as shown in screenshot! Is it possible to predict when y is in string format? I used onehot encoder but its not working.

Solution

This is not a regression task, but classification. 'not working' is not very informative, however normally you'd just map classes to integers. Either sklearn.preprocessing.LabelEncoder, sklearn.preprocessing.label_binarize().argmax(axis=1), pandas.factorize() or manual mapping should get the job done.

Worth noting support vector machines don't handle multiclass problems natively, so you may encounter troubles depending on the exact model you use. At least the latest sklearn versions should handle it automatically when using models like sklearn.svm.LinearSVC, building N binary classifiers under the hood.

I'd also recommend getting acquainted with a more elegant way of ensembling SVMs for multiclass problems, using sklearn.multiclass.OutputCodeClassifier().

Answered By - dx2-66