Majority-class baseline

The problem, from first principles

Before trusting any classifier, you need to know what it has to beat. A model that reaches 92 percent accuracy sounds strong until you notice that 92 percent of the labels were the same class. The majority-class baseline predicts the most frequent training label for every example. Any real model should clear it.

Implement majority_class_baseline(y_train, n_samples), which returns an array of length n_samples filled with the most frequent label in y_train.

Constraints

• When two labels tie for most frequent, return the smaller label (the first in sorted order).
• y_train must not be empty: raise ValueError.
• The returned array has the same dtype as y_train.
• Do not modify y_train.

Hints

Open one at a time. Each gives away a little more than the last.