plt-gp

CLI Reference

Overview

plt-gp reads HDF5 files produced by ChemGP (the Rust crate chemgp-core) and generates publication-quality PDF figures via chemparseplot.

Subcommand	Plot type	Backend
convergence	Oracle calls vs force/energy	plotnine
surface	2D PES contour + NEB path overlay	matplotlib
quality	GP surrogate progression (multi-panel)	matplotlib
rff	RFF vs exact GP error (two-panel)	plotnine
nll	MAP-NLL landscape in hyperparameter space	matplotlib
sensitivity	3x3 hyperparameter sensitivity grid	matplotlib
trust	Trust region illustration (1D slice)	matplotlib
variance	GP variance hatching on PES	matplotlib
fps	FPS subset PCA scatter	plotnine
profile	NEB energy profile across images	plotnine
landscape	2D NEB landscape (delegates to plt-neb)	matplotlib
batch	TOML-driven batch figure generation	(all)

Common Options

All single-plot subcommands accept:

--input / -i

Path to the HDF5 file (required)

--output / -o

Output PDF/PNG path (required)

--width

Figure width in inches (default varies by type)

--height

Figure height in inches (default varies by type)

--dpi

Resolution (default 300)

Single-Plot Usage

# Convergence curve from a minimize run
rgpycrumbs chemgp plt-gp convergence -i output/leps_minimize.h5 -o convergence.pdf

# 2D surface with NEB path overlay
rgpycrumbs chemgp plt-gp surface -i output/leps_neb.h5 -o surface.pdf

# NLL landscape
rgpycrumbs chemgp plt-gp nll -i output/leps_nll.h5 -o nll.pdf

Batch Mode

Define a TOML config listing all plots, then generate them in one command:

rgpycrumbs chemgp plt-gp batch --config figures.toml

See the batch tutorial for a worked example with a full TOML config.

Energy Clamping

Surface and quality plots auto-detect clamping ranges from the filename:

• Files containing mb use [-200, 50]

• Files containing leps use [-5, 5]

Override with --clamp-lo and --clamp-hi flags on the subcommand, or clamp_lo / clamp_hi keys in the TOML config.

API Reference

For developer details and source code see rgpycrumbs.chemgp.plt_gp.

The plotting functions themselves live in chemparseplot at chemparseplot.plot.chemgp.