Skip to content

Plot every pair of columns

pairs() draws a scatterplot matrix: every pair of a table's number columns against each other, one variable on x in each column of panels and one on y in each row. A variable's scale is shared along its column and its row, so a point sits at the same height across its row and at the same place across its column. The diagonal shows each variable on its own.

from anytrellis import pairs
from anytrellis.datasets import load

penguins = load("penguins")
plot = pairs(
    penguins,
    ["bill_length_mm", "bill_depth_mm", "flipper_length_mm", "body_mass_g"],
    color="species",
)
plot

The column strips on top and the row strips on the right name the variables; a panel's name is its row's variable, then its column's ("bill_depth_mm ยท bill_length_mm"), as panels() and open() use it.

Brush one panel, see the rows everywhere

Press s and drag a box in any panel: the rows inside it light up in every panel, so a cluster found in one pair of variables shows where it lies in all the others. Hovering a point shows its whole row.

selected() gives the brushed rows of your table, once each, as the same kind of table:

plot.selected()                 # a Polars DataFrame here: the brushed penguins

From Python, select in one panel, named by its row's and its column's variable: here the long, deep bills of bill depth (y) against bill length (x), mostly Chinstrap penguins:

plot.select(x=(45, 60), y=(17, 22), panel=("bill_depth_mm", "bill_length_mm"))
print(plot.selected()["species"].value_counts().sort("species"))
shape: (3, 2)
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚ species   โ”† count โ”‚
โ”‚ ---       โ”† ---   โ”‚
โ”‚ str       โ”† u32   โ”‚
โ•žโ•โ•โ•โ•โ•โ•โ•โ•โ•โ•โ•โ•ชโ•โ•โ•โ•โ•โ•โ•โ•ก
โ”‚ Adelie    โ”† 3     โ”‚
โ”‚ Chinstrap โ”† 57    โ”‚
โ”‚ Gentoo    โ”† 5     โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ดโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜

The diagonal

diag="auto" (the default) draws a density curve per colour when there is a colour column, and a histogram otherwise. Each is drawn over its variable's range, so the row's scale still reads for it: a curve or bar reaching the top is the most common value, not a count. Hover a curve or a bar for its density or count.

Argument Draws
diag="density" a density curve per colour, each weighted by its share of the rows
diag="hist" a histogram, stacked by colour; bins=20 sets its bins
diag=None nothing

Each pair once

corner=True keeps the panels below the diagonal (and the diagonal), since those above it show the same pairs mirrored:

plot = pairs(
    penguins,
    ["bill_length_mm", "bill_depth_mm", "flipper_length_mm"],
    color="species",
    diag="hist",
    corner=True,
)
plot

Options

Argument Sets
columns=[...] the number columns, in order (default: all of them but color)
color="species" a colour column (numbers: a colormap, the diagonal in grey) or one colour
palette=[...] colours for a category colour column
size=3, opacity=0.6 the points' diameter in pixels, and their opacity
tooltip=["island"] more columns shown on hover
title="Penguins" a title over the plot
panel_height=150, height=600 the panels' or the plot's height (default: square panels)
legend="bottom" the legend under the plot
renderer="webgl2" as for Trellis

A pairs() plot is a Trellis: it zooms (a panel's zoom moves its row and column), saves (save_html, save_png) and opens panels like any other. It draws its own layers, so it takes no more.

Large tables

Every row is drawn once in each panel off the diagonal: five columns of a million rows are 20 million points. Panels of more than 100,000 points are drawn as cells, as in Plot large data, and brushing still selects the rows themselves.