A scatter plot of 200,000 trips (distance vs. duration) is a solid black blob. What is this problem called?
Q2
Which remedy scales best to millions of points?
Q3
A 10 % random sample of 200 PM2.5 readings is used to "discover outliers". The sample contains none of the 4 outliers. What is the lesson?
Q4
What does aggregation (e.g., daily means of hourly readings) hide?
Q5
In a scatterplot matrix of 5 attributes, how many distinct pairwise scatter plots are there (ignoring the diagonal and mirrored copies)?
Q6
In parallel coordinates, what pattern between two adjacent axes indicates a strong negative correlation?
Q7
Why must attributes usually be standardized before PCA?
Q8
After standardization, PC1 explains 84.2 % and PC2 14.3 % of the variance. What does this mean for a 2D PCA plot?
Q9
What are PCA loadings?
Q10
Which statement about t-SNE plots is correct?
Q11
UMAP and t-SNE differ from PCA mainly because they are:
Q12
What does k-means minimize?
Q13
Inertia for k = 1, 2, 3, 4 is 32.0, 6.7, 2.3, 1.0. Which k does the elbow method suggest first?
Q14
Why is it a mistake to say "cluster 0 is the best stations" after k-means?
Q15
Why should k-means be run on standardized features?
Q16
Which reduction best supports "show the typical daily pattern and how variable it is" for a year of hourly PM2.5?
Q17
A 2D UMAP plot of 50,000 document embeddings shows two clusters far apart. A colleague concludes "these topics are completely unrelated". What is the problem?
Q18
When drawing parallel coordinates with thousands of polylines, which technique helps most?
Q19
A chart caption says only "Clusters of stations". What should it add for honest reporting?
Q20
A 2D histogram of 11 points on a 4 × 3 grid has 7 empty cells. What does a large share of empty cells suggest?
Q21
Which tool is specifically designed to render very large point datasets by aggregating them to the pixel grid?
Q22
For discrete values (e.g., ratings 1–5 on both axes), many points sit exactly on top of each other. Which simple remedy reveals how many there are?
Q23
Which statement about filtering as a reduction is correct?
Q24
Multidimensional scaling (MDS) primarily tries to preserve:
Q25
In t-SNE, what does the perplexity parameter roughly control?
Q26
A 2D PCA plot explains only 35 % of the variance. What should a careful analyst conclude?
Q27
A k-means analysis always returns exactly k clusters. Why is this a risk?
Q28Short answer
Interpret this PCA on the 8 CityPulse stations (standardized trips per day, mean duration, casual share, distance to centre): explained variance PC1 84.2 %, PC2 14.3 %; PC1 loadings +0.53, −0.52, −0.50, −0.45; PC2 loadings −0.21, −0.34, −0.54, +0.74. What do PC1 and PC2 mean, where would City Hall and the Airport appear, and what changes without standardization?
Q29Short answer
A data scientist shows a t-SNE map of 20,000 bike users (embedding of their trip histories) with five colourful clusters and claims: "There are exactly five rider types; cluster 3 is twice as large as cluster 1, and clusters 1 and 5 are the most different because they are farthest apart." Critique each claim and propose a better analysis.
Q30Short answer
CityPulse stores 1.5 billion sensor readings (60 sensors × every 10 seconds × 10 years). For each task, propose a reduction and say what it hides: (a) a city-wide 10-year trend; (b) find malfunctioning sensors; (c) an interactive scatter of temperature vs. PM2.5 for all readings; (d) investigate one pollution episode in detail.