API reference

corner(data, **kwargs)

One-shot convenience wrapper equivalent to Cornetto(data, ...).plot(...). Returns (fig, axes).

Analysis kwargs (Cornetto.__init__):

Name	Type	Default	Notes
data	dict[str, array]	(required)	1-D (N,) or 2-D (N_chains, N) arrays, per parameter.
params	list[str]	None	Subset to plot; unused arrays are never processed.
labels	dict[str, str]	{}	Display labels (LaTeX supported via raw strings).
truths	dict[str, scalar \| array]	None	Scalar shared; array length N_chains for per-chain truths.
chain_labels	list[str]	None	Legend labels, one per chain.
weights	array or list	None	Importance-sampling weights, one array per chain (or shared).
subsample	int	None	Thin each chain to this many samples.
ax_lims	dict[str, (lo, hi)]	None	Explicit axis ranges.
max_chains	int	10	Hard cap on simultaneous chains.
stat	str \| callable	"median"	See Statistics.
sigmas	tuple[float]	(1, 2)	n-sigma to 2-D level conversion.
kwargs_truths	dict	None	Truth line style (color, lw, ls, alpha, marker, markersize).
kwargs_stats	dict[str, dict]	None	Per-stat overrides (lw, ls, alpha).

Plot kwargs (Cornetto.plot):

Name	Default	Notes
color	"cornetto"	Named palette or list of hex codes.
dark	False	Dark theme.
smooth	True	KDE (True) or histogram (False) on the diagonal.
bandwidth	None	Explicit KDE bandwidth; overrides Silverman.
bandwidth_scale	0.8	Multiplier on Silverman's rule.
n_grid	128	KDE evaluation grid per axis.
fill_1d	True	Light tint under the 1-D curve.
fill_contours	True	Filled 2-D contour bands.
contour_lw	1.2	2-D contour line width.
show_tension	False	Overlap-integral annotation in 2-D panels.
annotate_peaks	False	Mark local maxima in 1-D marginals.
show_titles	True	value+err/-err heading (single-chain only).
figsize	None	Override figure size.
fig_size_per_dim	2.3	Inches per parameter when figsize is unset.
label_pad	0.08	hspace / wspace between panels.
title	None	Figure suptitle.
fig	None	Draw into an existing figure.
usetex	False	Use LaTeX for text rendering (requires a TeX install).
use_datashader	False	Rasterized scatter of raw samples in 2-D panels. Requires datashader pandas.

---

quick_corner(data, **kwargs)

Fast path - histograms only, same dict API.

Name	Default	Notes
data	(required)	As above.
params	None	Subset.
labels	{}
chain_labels	None
truths	None
weights	None
ax_lims	None
subsample	20_000	Default thin for responsiveness.
bins	30	Histogram bins per axis.
sigmas	(1, 2)
stat	"median"	Built-in name or callable; same as corner.
kwargs_stats	None	Per-stat {lw, ls, alpha} overrides.
color	"cornetto"
dark	False
figsize	None
fig_size_per_dim	2.0	Matches corner() for consistent output.
title	None
fig	None
usetex	False	Use LaTeX for text rendering.

Returns (fig, axes).

---

marginal(data, **kwargs)

1-D marginal distributions for all parameters in a grid layout.

Name	Default	Notes
data	(required)	Same dict API as corner.
params	None	Subset.
labels	{}
truths	None
chain_labels	None
weights	None
stat	"median"
ncols	5	Panels per row; last row padded with hidden axes.
color	"cornetto"
dark	False
show_titles	True	value+err/-err heading (single-chain only).
show_legend	None	Auto-show when n_chains >= 2.
fill_1d	False
annotate_peaks	False
figsize	None
fig_size_per_dim	2.5
title	None
usetex	False

Returns (fig, axes) where axes has shape (nrows, ncols).

---

trace(data, **kwargs)

Trace plot - one row per parameter, x = sample index, y = value. Uses datashader for rasterized rendering (falls back to matplotlib if unavailable).

Name	Default	Notes
data	(required)	Same dict API as corner.
params	None	Subset.
labels	{}
chain_labels	None
stride	50	Running-median window = max(1, N_samples // stride).
use_datashader	True	Rasterize via datashader; needs pip install datashader pandas.
datashader_kwargs	None	Dict with n_px_w, n_px_h, how, min_alpha, max_alpha.
color	"cornetto"
dark	False
show_legend	None	Auto-show when n_chains >= 2.
figsize	None
fig_height_per_param	1.5	Row height in inches.
title	None
usetex	False

Returns (fig, axes) where axes has shape (N_params,).

---

trace_marginal(data, **kwargs)

Combined 1-D KDE (left) + trace (right) - one row per parameter. The 1-D panel shows KDE curves without CI bands or titles.

Name	Default	Notes
data	(required)	Same dict API as corner.
params	None	Subset.
labels	{}
truths	None
chain_labels	None
stat	"median"
stride	50	Running-median window.
use_datashader	True	Rasterize traces via datashader.
datashader_kwargs	None	As for trace.
width_ratios	(1.0, 4.0)	Relative widths of the marginal and trace columns.
color	"cornetto"
dark	False
show_legend	None	Auto-show when n_chains >= 2.
figsize	None
fig_height_per_param	1.5	Row height in inches.
title	None
usetex	False

Returns (fig, axes) where axes has shape (N_params, 2). axes[:, 0] - 1-D marginal panels; axes[:, 1] - trace panels.

---

class Cornetto

Stateful interface - useful when you want to render multiple figures from the same dataset without recomputing KDEs, or to inspect/export statistics.

Cornetto(data, ...)

Same arguments as corner's analysis kwargs.

Cornetto.plot(...)

Same as corner's plot kwargs. KDEs and statistics are computed lazily and cached; subsequent calls with the same n_grid / bandwidth parameters reuse the cache.

Cornetto.marginal(...)

Same kwargs as marginal() (visual arguments only). Returns (fig, axes) with shape (nrows, ncols).

Cornetto.trace(...)

Same kwargs as trace(). Returns (fig, axes) with shape (N_params,).

Cornetto.trace_marginal(...)

Same kwargs as trace_marginal(). Returns (fig, axes) with shape (N_params, 2).

Cornetto.pairplot(other, ...)

Draw self in the lower triangle + diagonal and other in the upper triangle. other must be a Cornetto whose params are a subset of self's.

Cornetto.info()

Print the setup (stat, contour sigma mapping, truth style, palette, theme, KDE grid / bandwidth) used by the last plot() call. Intended as a readable replacement for legend clutter.

Cornetto.summary()

Return a SummaryTable with median, p16, p84, mean, std per (param, chain). Renders as HTML in Jupyter, plain text in the terminal.

Cornetto.latex(caption="", label="tbl:posterior", fmt=".4g", style="aastex")

Return a LaTeX table string via astropy.io.ascii. Requires astropy.

---

Built-in statistics

Name	Module	label
stat_median	cornetto	"median"
stat_median_mad	cornetto	"median±MAD"
stat_median_hdi	cornetto	"median+HDI68%"
stat_mean	cornetto	"mean±std"

Registry: cornetto.STAT_REGISTRY - {str: callable}.

Each function has the signature fn(samples, weights=None) -> dict and must return {"center": float, "lo": float, "hi": float, "label": str}.

---

Utilities

sigmas_to_levels(sigmas)

Convert n-sigma tuple to 2-D probability-mass levels via P(n) = 1 - exp(-n^2/2).

hdi(samples, prob=0.68)

Highest Density Interval - shortest interval containing prob of the mass.

compute_stats(samples, weights=None)

Returns {mean, median, std, p16, p84}.

Module constants

• cornetto.MAX_CHAINS - hard cap (10).
• cornetto.CORNETTO_PALETTE - multi-chain jewel-tone palette (10 colors).
• cornetto.PALETTES - named single-chain palettes.
• cornetto.DEFAULT_LEVELS - sigmas_to_levels((1, 2)).