Semi-supervised learning basics in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main advantage of semi-supervised learning?

Semi-supervised learning uses both labeled and unlabeled data. What is the main advantage of this approach compared to supervised learning?

A It does not need any data preprocessing.

B It always achieves higher accuracy than supervised learning.

C It only works with unlabeled data.

D It requires fewer labeled examples, reducing labeling cost.

💡 Hint

Think about the cost and effort of labeling data.

Semi-supervised learning reduces the need for many labeled examples by leveraging unlabeled data, which is usually easier and cheaper to collect.

❓ Predict Output

intermediate

Output of semi-supervised label propagation

Given the following Python code using label propagation, what is the predicted label for the unlabeled point?

ML Python

from sklearn.semi_supervised import LabelPropagation
import numpy as np

# Data points: 3 labeled, 1 unlabeled
X = np.array([[1, 2], [2, 3], [3, 4], [8, 9]])
# Labels: 0, 0, 1, -1 (unlabeled)
y = np.array([0, 0, 1, -1])

model = LabelPropagation()
model.fit(X, y)
predicted_label = model.transduction_[-1]

A Raises an error

B 0

C -1

D 1

💡 Hint

Label propagation assigns labels based on neighbors' labels.

The unlabeled point [8,9] is far from points labeled 0 and 1, but label propagation assigns it the label of the closest cluster, which is 0 in this case.

❓ Model Choice

advanced

Choosing a model for semi-supervised learning

You have a small labeled dataset and a large unlabeled dataset. Which model is best suited for semi-supervised learning in this scenario?

A Support Vector Machine with self-training

B Standard supervised Random Forest

C K-Means clustering

D Linear Regression

💡 Hint

Look for a model that can iteratively label unlabeled data.

Support Vector Machine with self-training can use labeled data to train, then label unlabeled data iteratively, making it suitable for semi-supervised learning.

❓ Hyperparameter

advanced

Key hyperparameter in label spreading

In label spreading, which hyperparameter controls how much the model trusts the initial labels versus the structure of the unlabeled data?

A Alpha (clamping factor)

B Learning rate

C Number of neighbors

D Regularization strength

💡 Hint

This parameter balances label retention and propagation.

Alpha controls the clamping factor: a higher alpha means more trust in the initial labels, while a lower alpha means more trust in the unlabeled data structure.

❓ Metrics

expert

Evaluating semi-supervised learning performance

You trained a semi-supervised model. Which metric is most appropriate to evaluate its performance on the labeled test set?

A Mean squared error

B Silhouette score

C Accuracy

D Adjusted Rand index

💡 Hint

Consider that you have true labels for the test set.

Accuracy measures how many test samples are correctly classified, which is suitable when true labels are available.