Get unnormalized predict_proba when using "ovr" in LogisticRegression

Viewed 26

When using the LogisitcRegression.predict_proba the docs say that, when multiclass = 'ovr' they return the normalized probability for each class.

Is there a way, without having to calculate it using the model.coef_ like

pred_proba_manually = [1/(1+np.exp(-(intercept + tf_idf_val@coef))) for
                      coef,intercept in zip(logreg.coef_,logreg.intercept_)]

to get the unnormalized probabilities for each class?

0 Answers
Related