One-vs-rest and one-vs-one strategies in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Difference between One-vs-Rest and One-vs-One

Which statement correctly describes the difference between the One-vs-Rest (OvR) and One-vs-One (OvO) strategies for multi-class classification?

A OvR trains classifiers for every pair of classes, while OvO trains one classifier per class against all others.

B OvR and OvO both train one classifier per class but use different loss functions.

C OvR uses a single classifier for all classes, while OvO uses multiple classifiers for each class.

D OvR trains one classifier per class against all other classes, while OvO trains classifiers for every pair of classes.

💡 Hint

Think about how many classifiers are trained in each strategy and what data they use.

One-vs-Rest trains one classifier per class, distinguishing that class from all others combined. One-vs-One trains a classifier for every pair of classes, focusing on distinguishing between just two classes at a time.

❓ Model Choice

intermediate

Choosing strategy for many classes

You have a dataset with 50 classes and limited computing power. Which multi-class classification strategy is generally more efficient to train?

A One-vs-Rest, because it trains fewer classifiers (one per class).

B One-vs-One, because it trains classifiers only on pairs of classes.

C Both strategies require the same training time regardless of class count.

D One-vs-One, because it requires less memory for large numbers of classes.

💡 Hint

Consider how the number of classifiers grows with the number of classes in each strategy.

One-vs-Rest trains one classifier per class, so with 50 classes it trains 50 classifiers. One-vs-One trains classifiers for every pair of classes, which is 50*49/2 = 1225 classifiers, much more computationally expensive.

❓ Predict Output

advanced

Output of OvR prediction voting

Given the following OvR classifiers' decision scores for a sample, what is the predicted class?

scores = {'class_A': -1.2, 'class_B': 0.5, 'class_C': 0.3}

OvR predicts the class with the highest decision score.

A class_B

B class_A

C class_C

D No prediction because scores are negative

💡 Hint

Look for the highest score regardless of sign.

OvR picks the class with the highest decision score. Here, class_B has the highest score 0.5, so it is predicted.

❓ Metrics

advanced

Calculating number of classifiers in OvO

For a multi-class problem with 7 classes, how many binary classifiers does the One-vs-One strategy train?

A 49

B 21

C 42

D 7

💡 Hint

Use the formula for combinations of 7 classes taken 2 at a time.

Number of classifiers in OvO is n*(n-1)/2 = 7*6/2 = 21.

🔧 Debug

expert

Error in OvR prediction code

What error will this Python code raise?

def predict_ovr(scores):
    # scores is a dict of class:score
    max_score = max(scores.values())
    for cls, score in scores.items():
        if score == max_score:
            return cls

result = predict_ovr({'A': 0.2, 'B': 0.2, 'C': 0.1})
print(result)

A Raises ValueError due to multiple max values.

B Returns 'B' because it finds the last max score match.

C Returns 'A' because it finds the first max score match.

D Raises TypeError because max() cannot be used on dict values.

💡 Hint

Consider how max() and the for loop behave with ties.

The code returns the first class with the max score. Since 'A' and 'B' tie at 0.2, it returns 'A'. No error is raised.