Quickstart¶

Install¶

pip install cornetto

Your first corner plot¶

Cornetto takes a dict[str, array]: parameter name to 1-D sample array.

# %%
import numpy as np
from cornetto import corner

rng = np.random.default_rng(0)
data = {
    "m1":  rng.normal(30, 4, 10_000),
    "m2":  rng.normal(25, 3, 10_000),
    "chi": rng.uniform(-1, 1, 10_000),
}

fig, axes = corner(data)

That's it. Labels default to the dict keys; contour levels default to 1σ and 2σ (39.3% and 86.5% in the 2-D sense, see Statistics).

LaTeX labels¶

# %%
corner(data, labels={
    "m1":  r"$m_1\,[M_\odot]$",
    "m2":  r"$m_2\,[M_\odot]$",
    "chi": r"$\chi_{\mathrm{eff}}$",
})

Truth markers¶

# %%
corner(data, truths={"m1": 30.0, "chi": 0.0})

A scalar applies to every panel for that parameter. Pass an array of length N_chains to set a different truth for each chain.

Don't need dotted lines everywhere?

Set kwargs_truths={"ls": "-", "marker": "o"} for solid crosshairs plus a dot at the truth location, or pass kwargs_truths={"alpha": 0.0} to hide the lines entirely.

Subset of parameters¶

# %%
corner(data, params=["m1", "m2"])   # only plot these two

Cornetto only processes the arrays it actually draws, so this is safe to use on wide chains.

Saving the figure¶

corner() returns a standard (fig, axes) pair. Save it like any matplotlib figure. PDF or SVG keeps vector quality for papers and talks.

# %%
fig, axes = corner(data)
fig.savefig("posterior.pdf", bbox_inches="tight")   # vector, scales cleanly
# or: fig.savefig("posterior.svg", bbox_inches="tight")

What the legend shows¶

Only the chain name(s). Everything else (statistic, contour levels, truth styling) is available through Cornetto.info(), so the plot stays clean.

# %%
from cornetto import Cornetto
c = Cornetto(data, chain_labels=["mock"])
fig, axes = c.plot()
c.info()

Cornetto setup
──────────────
  params         : ['m1', 'm2', 'chi']
  n_chains       : 1
  stat           : median  (center line + shaded interval on 1-D marginals)
  contour sigmas : 1σ→39.3%, 2σ→86.5%
  …

Next steps¶

User guide — multi-event, weights, statistics, styling.
Fast mode — sub-second plots for iteration.