Chap 3, more clarification for Never5Classifier(BaseEstimator)

[hypntzed78]

from sklearn.base import BaseEstimator
class Never5Classifier(BaseEstimator):
    def fit(self, X, y=None):
        pass
    def predict(self, X):
        return np.zeros((len(X), 1), dtype=bool)

I don't understand how this being created. I thought it would predict not-5 class is true, but whereas y_train[1] == 0 , never_5_clf.predict(X_train) is still False for the [1] ? plz explain how this classifier work. Which code represent it can predict the not-5?

[ageron]

Thanks for your question.

The Never5Classifier always predicts False, no matter what you give it as input. The goal is to point out that even such a dumb classifier gets over 90% accuracy, and that's just because the dataset is skewed: only 10% of the instances are 5s, so 90% of them are not 5s.

The instance X_train[1] represents a zero (as you can see because y_train[1] == 0). So it's not a 5. Therefore the Never5Classifier is correct when it predicts False, meaning "this is not a 5". Despite being so dumb, the classifier will get it right roughly 90% of the time.

Hope this helps.

[hypntzed78]

Thanks for your question.

The Never5Classifier always predicts False, no matter what you give it as input. The goal is to point out that even such a dumb classifier gets over 90% accuracy, and that's just because the dataset is skewed: only 10% of the instances are 5s, so 90% of them are not 5s.

The instance X_train[1] represents a zero (as you can see because y_train[1] == 0). So it's not a 5. Therefore the Never5Classifier is correct when it predicts False, meaning "this is not a 5". Despite being so dumb, the classifier will get it right roughly 90% of the time.

Hope this helps.

@ageron thank you so much, I understand your idea.