Fast mode quick_corner¶
When you're tweaking priors, comparing chains, or sanity-checking a new sampler run, slow renders get in the way. quick_corner uses the same API as corner but skips KDE - histograms only.
API parity¶
# %%
from cornetto import quick_corner
fig, axes = quick_corner(
data,
chain_labels=["my chain"],
truths={"m1": 30.0},
labels={"m1": r"$m_1$"},
)
All the same semantics as corner:
- dict-of-arrays input, 1-D or 2-D
params,labels,truths,chain_labels,weights,ax_limsstat,kwargs_stats- full custom statistic, same built-ins ("median","median_mad","median_hdi","mean") or your own callablecolor,darksigmasfor contour levels- returns
(fig, axes)with the same outer dimensions ascorner()
What it does differently¶
| Feature | corner |
quick_corner |
|---|---|---|
| 1-D marginals | FFT-KDE | Histogram (step line) |
| 2-D marginals | FFT-KDE + contours | histogram2d + contours |
Default subsample |
none | 20 000 |
Beyond that, panels are drawn lighter: no fills, titles, peak markers, or additional annotations.
Why it's fast¶
- No KDE. FFT-KDE on an 8-parameter corner means 28 two-dimensional KDE
calls - the dominant cost in
corner().quick_cornerreplaces them with histograms. - Minimal drawing. Coarse histogram grid, one contour line per sigma, no peak detection, no tension annotation, no titles.
See the Performance page for numbers.