Majority Vote

Combine the class predictions of several models by majority vote.

An ensemble of classifiers often beats each of its members, because their errors do not all happen on the same examples. The simplest way to combine them is hard voting: each model casts one vote per example, and the label with the most votes wins.

You get a (n_models, n_samples) array of integer labels. Return, for each sample, the label that appears most often in its column. When two or more labels tie, return the smallest of them.

Function signature

def majority_vote(predictions):

Example

Input

predictions = np.array([
    [0, 1, 2, 1],
    [0, 2, 2, 0],
    [1, 1, 0, 0],
])
majority_vote(predictions)

Output

array([0, 1, 2, 0])

Reasoning

Read each column from top to bottom. Column 0 has votes 0, 0, 1, so 0 wins. Column 1 has 1, 2, 1, so 1 wins. Column 2 has 2, 2, 0, so 2 wins. Column 3 has 1, 0, 0, so 0 wins.

Constraints and notes

• Labels are non-negative integers.
• Break ties by the smallest label.
• Return an integer array of shape (n_samples,).