dysonsphere

pip install dysonsphere

import altair as alt
import numpy as np
import polars as pl
import dysonsphere as ds

ds.theme(palette="blues2", chartWidth=140)

rng = np.random.default_rng(7)
rep1 = rng.normal(10, 2, 2500)
df = pl.DataFrame({"rep1": rep1, "rep2": rep1 * 0.9 + rng.normal(1, 0.9, 2500)})

heatmap = (
    alt.Chart(df)
    .mark_rect()
    .encode(
        x=alt.X("rep1:Q", bin=alt.Bin(maxbins=24), title="Replicate 1"),
        y=alt.Y("rep2:Q", bin=alt.Bin(maxbins=24), title="Replicate 2"),
        color=alt.Color("count():Q", title=None),
    )
)

chart = heatmap + ds.add_correlation(df, "rep1", "rep2")

Publication defaults

Charts are authored at print scale with considered typography, tick geometry, and mark styling - the SVG that comes out is the figure that goes in the paper.

Perceptually uniform palettes

300+ palettes built in Oklab and resampled by arc length, so sequential ramps step evenly and diverging ramps balance. Exportable straight to Illustrator swatches.

Statistical annotations

Omnibus tests, post-hoc brackets, correlation readouts - computed with scipy, drawn as layers you compose with +, reported with exact p-values and effect sizes.

Composable marks & layers

Violins, strip plots, reference lines, text, shading, point labels, multilabels, log/power minor ticks - all native Altair objects that layer onto your own charts.

Self-describing exports

save() embeds provenance, statistics, the theme, and your data into SVG, PNG, and JSON; read() and load() pull them back out. Every figure is its own lab notebook entry.

Runs in your browser

The Chart Studio runs the real library via WebAssembly - upload your own data or write Python against it, no install, same output as your notebook.