NCA-ADS - Machine Learning With RAPIDS (16% of the exam) - Section 2.2
Apply regression, classification, and clustering.
Distinguish between regression, classification, and clustering as supervised and unsupervised problem types. Select and apply the appropriate algorithm family based on whether the target variable is continuous, categorical, or absent.
Practice question for this objective
lock_openFree sampleMachine Learning With RAPIDSmedium
A data scientist needs to predict the exact sale price of a house given its features such as square footage, number of bedrooms, and location. Which cuML algorithm family and target type correctly describe this task?
AClassification, because the model assigns each house to a price category
BClustering, because grouping similar houses reveals natural price segments without needing labelled training data
CRegression, because the target variable is a continuous numeric value and the model learns from labelled training examplescheck_circle Correct
DDimensionality reduction, because feature engineering must precede any numeric prediction task
Distinguish supervised regression from classification and unsupervised clustering by identifying whether the target is continuous, discrete, or absent. Supervised learning requires labelled training data. When the label is a continuous numeric quantity - such as a sale price - the task is regression. cuML's linear regression and random forest regressor both operate on continuous targets and run natively on GPU memory via cuDF DataFrames. Classification, by contrast, maps inputs to a finite discrete label set (for example, price tier A/B/C). Clustering such as k-means or DBSCAN operates without labels and groups observations by similarity rather than predicting a numeric output for each sample.
Why A is wrong: Classification predicts a discrete label from a finite set of categories, not a continuous numeric value such as a sale price expressed in dollars
Why B is wrong: Clustering is unsupervised and produces group assignments rather than a numeric prediction for each individual observation; labelled sale prices already exist in the training set
Why C is correct: Regression is the supervised learning family for continuous targets. Sale price is a continuous numeric output, and cuML provides GPU-accelerated estimators such as linear regression and random forest regressor that fit this exact task
Why D is wrong: Dimensionality reduction is a data preprocessing technique, not a predictive modelling family; it does not produce per-sample numeric predictions and does not replace the choice of regression
Answers that look right on this material and are not. Each one is a distractor from a different question in the NCA-ADS bank for this domain.
cuML logistic regression, trained on the transaction records to assign each customer to a spending class
errorWhy it is wrong: Logistic regression is a supervised classification algorithm that requires labelled training examples; no class labels exist in this scenario, so it cannot be applied without first inventing artificial target columns
The cuML linear regression estimator, because it groups rows into clusters by fitting a line that minimises the squared distance from each point to its assigned cluster centre.
errorWhy it is wrong: Tempting because it mentions minimising squared distance, but it is wrong: linear regression is a supervised method that predicts a continuous target, it does not partition unlabelled rows into clusters.
cuML KMeans, because it scales to hundreds of thousands of rows on the GPU and assigns every point to exactly one centroid-based cluster.
errorWhy it is wrong: KMeans does scale well on the GPU, which makes it tempting, but it requires the cluster count up front, assumes roughly spherical equal-variance clusters, and forces every point into a cluster, so it cannot leave stray pings as noise or capture elongated road-shaped groups.