`profile_2d`

Render a 2D profile-likelihood (or profile-χ²) map: the samples are reduced onto a grid by taking the best value per bin, then drawn as a heatmap, optionally with credible-region contours. This is the standard frequentist 2D panel.

It is an encapsulated figure type: set type: profile_2d on a figure and Jarvis-PLOT builds the layers for you.

What it expands to

You make three orthogonal choices — how to reduce, how to render, and whether to draw credible regions:

Layer 1  profile transform → reduced points
Layer 2  pcolormesh (interp: true or method: grid)  OR  voronoi cells (interp: false)
Layer 3  contour            ← credible region, if requested
+ extra_layers (yours)

Required keys

Key	Meaning
`data`	DataSet name (or list → concatenated)
`x`, `y`	Axis coordinates `{expr, lim, scale, label}`
`z`	The quantity to profile, e.g. `{expr: LogL, label: "..."}`

Type-specific keys

Key	Default	Purpose
`method`	`bridson`	Reduction grid: `bridson` (quasi-uniform) or `grid` (uniform bins)
`bins`	`100`	Reduction resolution
`objective`	`max`	Per-bin reducer: `max`, `min`, `mean`
`interp`	`true`	`true` → smooth `pcolormesh`; `false` → `voronoi` cells
`grid`	`500`	Interpolation grid size (when `interp: true`)
`grid_points`	`rect`	Reduction geometry: `rect` or `ternary`
`seed`	`null`	RNG seed (for `bridson`)
`credible_region`	off	Credible-region contours (see below)
`colorbar`	see common	Default label = the `z` label, `vmin` = `auto`

Plus the common keys: name, style_card, frame, extra_layers.

ℹ️ Reduce vs. render method chooses how points are reduced (bridson/grid); interp chooses how the result is rendered (smooth pcolormesh vs. voronoi cells). They are independent — e.g. method: bridson + interp: false draws Voronoi cells over quasi-uniform support points.

`credible_region`

Off by default. Provide a block to overlay contours. Choose either statistical sigma levels or explicit levels:

credible_region:
  sigma: [1, 2]        # σ levels → profile-likelihood contours
  ndof: 2              # chi-square degrees of freedom (default 2)
  colors: [black, white]
  linewidths: [0.2, 0.2]

credible_region:
  levels: [-10, -5]    # explicit z levels instead of sigma
  colors: [black]

Examples

Minimal

Defaults to bridson reduction with smooth interpolation; a colorbar appears automatically.

- name: XY_profLogL
  type: profile_2d
  data: my_samples
  x: {expr: xx, lim: [0.1, 5], scale: log, label: "$x$"}
  y: {expr: yy, lim: [0, 5], label: "$y$"}
  z: {expr: LogL, label: "$\\log\\mathcal{L}$"}

Full — explicit reduction, credible region, custom colorbar

- name: XY_profLogL
  type: profile_2d
  data: [df_samples_0, df_samples_1]
  x: {expr: xx, lim: [0.1, 5], scale: log, label: "$x$"}
  y: {expr: yy, lim: [0, 5], label: "$y$"}
  z: {expr: LogL, label: "$\\log\\mathcal{L}$"}

  method: bridson
  bins: 150
  objective: max
  interp: true
  grid: 500

  colorbar: {cmap: jarvis_rainbow2_r, vmin: -50, vmax: 0}

  credible_region:
    sigma: [1, 2]
    colors: [black, white]
    linewidths: [0.3, 0.3]
    ndof: 2

Voronoi-cell rendering

- name: XY_profLogL_cells
  type: profile_2d
  data: my_samples
  x: {expr: xx, lim: [0, 1.4], label: "$x$"}
  y: {expr: yy, lim: [0, 1.4], label: "$y$"}
  z: {expr: LogL, label: "$\\log\\mathcal{L}$"}
  interp: false            # render reduced points directly as Voronoi cells

Notes

Use objective: max for a log-likelihood (profile the maximum), objective: min for a χ² (profile the minimum).
sigma + ndof produce proper profile-likelihood thresholds; use levels when you want exact z values instead.
To see the exact layers this expands to, run with --debug.