Machine Learning Methods: 10 practice questions

Question 1 of 10

What defines a supervised learning problem?

1. AThe inputs omit targets so the system must discover their internal groupings
2. BAn agent receives rewards while interacting with a changing environment
3. CThe model selects actions by balancing uncertain and known rewards
4. DThe training inputs include target values that provide known outcomes

Question 2 of 10

What defines an unsupervised learning problem?

1. AThe inputs have no labels or targets, so outputs rely on discovered patterns
2. BA model predicts one of several predefined classes from correctly labeled examples
3. CAn agent takes trial-and-error actions to maximize long-term rewards
4. DEvery input has a categorical or continuous target supplied during training

Question 3 of 10

What is the central learning mechanism in reinforcement learning?

1. AA regression model estimates a continuous target from related variables
2. BA system groups unlabeled examples once without receiving environmental feedback
3. CA model maps labeled input attributes directly to supplied target values
4. DAn agent learns through trial-and-error interaction to maximize long-term rewards

Question 4 of 10

Within supervised learning, when is a problem classification rather than regression?

1. AThe inputs have no targets and must be divided into discovered clusters
2. BThe agent receives rewards after actions in a dynamic environment
3. CThe target is categorical, such as one of several predefined classes
4. DThe target is continuous, such as an estimated house sale price

Question 5 of 10

Which task is a supervised regression problem?

1. ADiscovering customer groups from attributes when no group labels exist
2. BAssigning labeled documents to one of several predefined topic classes
3. CEstimating a house price from features when historical sale prices are supplied
4. DLearning a driving policy from actions and long-term track rewards

Question 6 of 10

What is the goal of cluster analysis in unsupervised learning?

1. AMaximize long-term rewards by selecting actions in a dynamic environment
2. BEstimate a continuous target value from features paired with known answers
3. CForm groups whose members are similar within a group and different across groups
4. DChoose between two predefined classes from correctly labeled examples

Question 7 of 10

How does reinforcement learning balance exploration and exploitation?

1. AIt trades labeled rows against unlabeled rows to create fixed cluster names
2. BIt trades training features against target columns to remove environmental feedback
3. CIt trades off actions with uncertain rewards against actions with known rewards
4. DIt trades categorical targets against continuous targets in the same regression output

Question 8 of 10

A retailer holds historical records in which each sale carries a product category and the amount it sold for. One workflow must predict the category of a new item, another must predict the amount itself. Which approach fits both?

1. AUse supervised regression for both, coding each category as a number so one model covers them
2. BUse supervised classification for the category and supervised regression for the price
3. CUse supervised classification for both, by grouping the amounts into price bands first
4. DUse supervised regression for the category and supervised classification for the amount

Question 9 of 10

A fraud team has transaction features but no fraud labels. It wants to discover natural groupings and flag rare observations that differ sharply from the rest. Which paradigm best matches both goals?

1. AUnsupervised learning, using clustering and anomaly detection on the unlabeled inputs
2. BSupervised regression, predicting the transaction amount so unusual amounts stand out
3. CSupervised learning, because anomaly detection needs a labeled example of each fraud pattern
4. DReinforcement learning, because flagging a transaction and learning the outcome is trial and error

Question 10 of 10

A warehouse robot has no record of the correct move for any situation. It tries a whole route, and the only feedback is a score once the run ends, which it must learn to improve over many runs. Which paradigm should be used?

1. AReinforcement learning, because the agent learns by interaction while maximizing long-term reward
2. BUnsupervised learning, because without a correct move recorded the robot only holds unlabeled inputs
3. CSupervised classification, labeling each finished route as a good route or a bad one
4. DSupervised regression, using the end-of-run score as the continuous target to predict