Regression testing in Software Engineering - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Purpose of Regression Testing in ML Models

Why is regression testing important when updating a machine learning model?

A To ensure new changes do not break existing model predictions

B To increase the training speed of the model

C To reduce the size of the training dataset

D To improve the model's ability to learn new features

💡 Hint

Think about what could happen if changes cause unexpected errors.

Regression testing checks that updates or fixes do not cause old features or predictions to fail.

❓ Metrics

intermediate

Detecting Regression with Metrics

Which metric change would best indicate a regression in a regression model's performance after an update?

A Increase in Mean Squared Error (MSE)

B Increase in training speed

C Decrease in model size

D Increase in number of features

💡 Hint

Think about what a higher error means for model predictions.

An increase in MSE means the model's predictions are less accurate, indicating regression.

🔍 Analysis

advanced

Identifying Regression Cause in Code

After retraining a model, predictions are worse. Which code change is most likely causing regression?

Software Engineering

old code:
model.fit(X_train, y_train)
new code:
model.fit(X_train, y_train, epochs=1)

A Increasing epochs causes overfitting

B Changing the dataset causes regression

C Reducing training epochs to 1 causes underfitting

D Using a different optimizer causes regression

💡 Hint

Think about how training time affects learning.

Training for only 1 epoch may not be enough for the model to learn patterns, causing worse predictions.

❓ Model Choice

advanced

Choosing a Model for Regression Testing

Which model type is best suited for regression testing to detect performance drops on continuous output tasks?

A K-Means Clustering

B Linear Regression

C Decision Tree Classifier

D Naive Bayes Classifier

💡 Hint

Consider the type of output the model predicts.

Linear Regression predicts continuous values, making it suitable for regression testing on continuous outputs.

🔍 Analysis

expert

Output of Regression Test Code Snippet

What is the output of this Python code simulating a regression test on model predictions?

Software Engineering

old_preds = [2.5, 0.0, 2.1, 7.8]
new_preds = [2.7, 0.1, 2.0, 7.9]
threshold = 0.15
regression_detected = any(abs(o - n) > threshold for o, n in zip(old_preds, new_preds))
print(regression_detected)

A TypeError

B False

C SyntaxError

D True

💡 Hint

Check if any prediction difference exceeds the threshold.

The difference between 2.5 and 2.7 is 0.2, which is greater than 0.15, so regression_detected is True.

Practice

1. What is the main purpose of regression testing in software development?

easy

A. To check that new code changes do not break existing features

B. To test the software performance under heavy load

C. To find security vulnerabilities in the software

D. To design the user interface of the application

Solution

1. Step 1: Understand regression testing goal

   Regression testing is done to ensure that recent code changes do not negatively affect existing functionality.

2. Step 2: Match purpose with options

   To check that new code changes do not break existing features clearly states this purpose, while others focus on performance, security, or design, which are different testing types.

3. Final Answer:

   To check that new code changes do not break existing features -> Option A

4. Quick Check:

   Regression testing = check existing features still work [OK]

Hint: Regression testing = verify old features after changes [OK]

Common Mistakes:

• Confusing regression testing with performance testing
• Thinking regression testing finds new bugs only
• Mixing regression testing with security testing

2. Which of the following is the correct way to describe an automated regression test?

easy

A. A script that runs tests automatically after code changes

B. A report summarizing user feedback

C. A manual checklist to verify UI changes

D. A design document for new features

Solution

1. Step 1: Identify automation in regression testing

   Automated regression tests are scripts that run automatically to check software after changes.

2. Step 2: Compare options to automation definition

   A script that runs tests automatically after code changes matches this exactly. Others describe manual or unrelated activities.

3. Final Answer:

   A script that runs tests automatically after code changes -> Option A

4. Quick Check:

   Automated regression test = automatic script [OK]

Hint: Automation means tests run by script, not manually [OK]

Common Mistakes:

• Confusing manual checks with automated tests
• Thinking reports or documents are tests
• Ignoring the 'automatic' part of regression testing

3. Consider this Python code snippet for a regression test automation check:

def test_model_accuracy(model, test_data, expected_accuracy):
    accuracy = model.evaluate(test_data)
    return accuracy >= expected_accuracy

class DummyModel:
    def evaluate(self, data):
        return 0.85

model = DummyModel()
result = test_model_accuracy(model, None, 0.80)
print(result)

What will be the output when this code runs?

medium

A. Error: 'NoneType' object is not iterable

B. False

C. True

D. 0.85

Solution

1. Step 1: Understand the evaluate method output

   The DummyModel's evaluate method returns 0.85 regardless of input.

2. Step 2: Check the comparison in test_model_accuracy

   It compares 0.85 >= 0.80, which is True, so the function returns True.

3. Final Answer:

   True -> Option C

4. Quick Check:

   0.85 >= 0.80 = True [OK]

Hint: Check returned value and compare to expected threshold [OK]

Common Mistakes:

• Assuming None input causes error
• Confusing returned accuracy with boolean result
• Ignoring the fixed return value of evaluate

4. You have this regression test code snippet in Python:

def run_regression_tests(tests):
    for test in tests:
        if not test():
            print("Test failed")
            break
    else:
        print("All tests passed")

def test1():
    return True

def test2():
    return False

run_regression_tests([test1, test2])

What is the output of this code?

medium

A. All tests passed

B. Test failed

C. No output

D. Error: break outside loop

Solution

1. Step 1: Analyze the loop over tests

   The loop runs test1() which returns True, so continues to test2().

2. Step 2: Check test2() result and loop behavior

   test2() returns False, triggering print("Test failed") and break to exit loop.

3. Final Answer:

   Test failed -> Option B

4. Quick Check:

   Second test fails, loop breaks, prints 'Test failed' [OK]

Hint: Loop breaks on first failed test, prints failure [OK]

Common Mistakes:

• Thinking else after for runs always
• Ignoring break statement effect
• Assuming all tests run regardless of failure

5. You want to improve your regression testing for a machine learning model that predicts house prices. Which approach best ensures your tests catch errors after model updates?

hard

A. Skip testing if training loss decreases

B. Manually check a few random predictions after each update

C. Only test the model when users report errors

D. Run automated tests comparing model predictions on a fixed test set before and after updates

Solution

1. Step 1: Identify best regression testing practice for ML models

   Automated tests on a fixed test set ensure consistent checks after updates.

2. Step 2: Evaluate options for reliability

   Run automated tests comparing model predictions on a fixed test set before and after updates uses automation and fixed data, catching errors reliably. Others rely on manual or reactive methods, which are less reliable.

3. Final Answer:

   Run automated tests comparing model predictions on a fixed test set before and after updates -> Option D

4. Quick Check:

   Automated fixed tests catch errors best [OK]

Hint: Use automated tests on fixed data to catch changes [OK]

Common Mistakes:

• Relying on manual checks only
• Ignoring testing until errors reported
• Assuming lower training loss means no bugs