corrheatmap() — Correlation heatmap

A coloured grid of how each pair of variables relates.

heatmap(cor(data))

Key idea

• A heatmap is a grid of cells where colour stands for a value.
• Darker or warmer cells usually mean higher numbers. (e.g. 0.9 dark, 0.1 light)
• A correlation heatmap shows how strongly each pair of variables moves together; correlation runs from -1 to +1. (e.g. +1 strong, 0 none, -1 opposite)
• It quickly highlights clusters and patterns across many variables at once.
• The diagonal is always the strongest because each variable matches itself perfectly. (e.g. value = 1.0)

Definition

$$r_{ij}=\frac{\mathrm{cov}(X_i,X_j)}{s_i{s}_j}$$

$X_i,X_j$

two variables (a row and a column)

$\mathrm{cov}(X_i,X_j)$

how they vary together

$s_i,s_j$

their standard deviations

$r_{ij}$

the correlation in that cell (-1 to +1)

In plain words

A coloured table where warm squares are "friends" that rise and fall together, and cool squares move in opposite directions.

Where you'd use it

• You have grades in math, physics, and art. A corrheatmap colours each pair’s correlation, so a warm square between math and physics (say +0.8) shows they rise together while a cool art–math square reveals little link.
• Before building a prediction model, a heatmap of all your features lets you scan dozens of relationships at once and pick out which inputs are strongly tied to the target.
• If two columns show a near +1 warm square, the heatmap warns they basically duplicate each other, so you can drop one to simplify the model.

Frequently asked questions

What does the heatmap colour show?

Pairwise correlations. Centre the colour scale at 0 so + and − read clearly.