planviz

The figure library for OpenPlan Labs — grid maps, search animations, plan timelines and benchmark charts, in the Frontier palette, light and dark from the same call.

Planners not included. This draws results; it does not produce them. It knows nothing about any solver's types: a grid is a 2-D array where truthy means blocked, a path is a sequence of (row, col) pairs, and a benchmark series is a mapping. Anything that can produce those can be plotted.

That figure is the reason this package exists. Every search algorithm has the same three sets — the nodes you expanded, the nodes on the frontier, and the path you returned — so the brand's three colours are that legend, and the three sets differ by shape as well as hue, which is what survives greyscale printing and red/green colour blindness. Drawing it correctly by hand, in four repositories, twice each for light and dark, was the duplication this replaces.

Install

pip install planviz

Two dependencies, both of which a plotting library obviously needs: matplotlib>=3.7 and numpy>=1.24. Python 3.10 or newer.

Not yet on PyPI — until then: pip install git+https://github.com/openplan-labs/planviz.

GIF and MP4 writing needs a little more:

pip install 'planviz[animation]'   # pillow for GIF, a bundled ffmpeg for MP4

Quickstart

import planviz

grid = [[0, 0, 0, 0], [0, 1, 1, 0], [0, 0, 0, 0]]   # truthy = blocked

ax = planviz.draw_search(
    expanded=[(0, 0), (1, 0), (2, 0), (2, 1)],
    frontier=[(0, 1), (2, 2)],
    path=[(0, 0), (1, 0), (2, 0), (2, 1), (2, 2)],
    grid=grid,
    dark=True,
)
planviz.save(ax, "search.png")

Four things hold across the whole API:

• Importing planviz changes no matplotlib state. The style is applied by planviz.use_style(dark=...), or per-figure inside a style_context that restores rcParams on exit. A solver library can depend on this without repainting its user's notebook.
• Every figure function takes dark: bool = False and an optional ax=, so light and dark variants come from one call and figures compose into panels.
• Nothing is saved or shown for you. Functions return the Axes they drew on (or the Figure, for multi-panel figures). planviz.save(...) is the explicit write.
• The brand ships inside the wheel. planviz/tokens.py and planviz/styles/frontier.mplstyle are generated from openplan-labs/branding and checked for drift in CI, so nothing looks up a repository at runtime and a figure rendered offline matches one rendered on a laptop.

To style figures you draw yourself with the same values:

import matplotlib.pyplot as plt
from planviz import tokens

with planviz.style_context(dark=False) as t:
    fig, ax = plt.subplots()
    ax.plot(xs, ys, color=t.path)          # the solution: the only warm value
    ax.plot(xs, others, color=t.agent(0))  # supporting series: the agent ramp

Gallery

Every image below is generated by examples/gallery.py from synthetic data, in both schemes, and is regenerated in CI. The full gallery with code for each figure is at openplan-labs.github.io/planviz/gallery.

Grids and agents

draw_grid	draw_paths
draw_paths(highlight=...)	draw_heatmap

animate_search and animate_paths. The frontier moves, the expanded set accumulates, and the path appears once at the end and stays — the accumulated closed list is the cost of the search, so erasing it hides the thing the figure is arguing about. GIFs are capped at 12 fps and 800 px wide, because they are read in a README on a train.

Search progress

search_panels	radial_wavefront
plan_timeline	plan_timeline(timeline_from_paths(...))

Benchmarks

scaling_curve	success_heatmap
phase_breakdown	crossover_plot

The API

Full signatures: API reference.

What it does not do

• No solver adapters. There is no plot_solution(pymapf.Solution). Callers pass arrays and mappings, which is what keeps one library serving four repositories with different problem types.
• No grouped-bar chart and no parity scatter yet. pymapf's plot_cost_comparison and cuplan's _fig_quality have no home here in 1.0.0; see migration.
• No 3-D space-time cube. pymapf.viz.plot_spacetime stays where it is.
• No live views. LiveSolveView and jupyddl's LiveSearchPlot are solver observers, not figures; they belong with the solver they observe.
• No interactivity, no web renderer. Static matplotlib output, plus GIF and MP4.

Who uses this

docs/migration.md maps each of their existing functions onto a planviz call, one line at a time.

Contributing

Bug reports, figures that are wrong, and figures that are missing are all welcome — see CONTRIBUTING.md. The binding constraint is brand/figures.md; the parts of it this library enforces are summarised in design rules.

MIT licensed.