Visualising big data
Showing more points than the screen has pixels: overplotting and its remedies (transparency, sampling, bins, small multiples, logarithmic scale), aggregating before display and the overview, zoom, details rule.
- 20 cards
- 6 topics
- Free
The outline, topic by topic
Each topic is weighted the way the exam weighs it.
- Concepts
- Questions
- Sort and link
- The numbers
- Scenarios
- Understand
A few cards from the deck
Overplotting: what is it?
Show the answer
A chart defect where so many points overlap that neither their density nor their structure can be told apart.
With 2 million points in a 320 × 200 pixel area, each pixel gets about 31 on average: the blob is solid and hides clusters.
A scatter plot of 2 million points is a solid blob. Which remedy reduces overplotting without losing the clusters?
- Bin them and colour each bin by its count
- Enlarge each point so it is easier to see
- Draw only the first 1,000 points of the file
- Shrink each point to a single pixel
Show the answer
Bin them and colour each bin by its count
Bigger points overlap more. The first thousand points are a biased sample. Even shrunk to one pixel, millions of points stack dozens per pixel: the blob stays solid. Bins count points and reveal density.
On a … scale, each tick multiplies the value by the same factor (say 10) instead of adding the same amount.
Show the answer
On a logarithmic scale, each tick multiplies the value by the same factor (say 10) instead of adding the same amount.
It makes values spanning several orders of magnitude readable, such as 10, 1,000 and 100,000 on the same axis. Zero does not exist on it.
Binning: what is it?
Show the answer
Splitting the chart area into cells whose points are counted, so only those counts are displayed.
The number of cells is set by the screen, not by the data: 2 million points or 2 billion yield the same number of cells to draw.