Ultraplot

UltraPlot

UltraPlot is an object-oriented superset of matplotlib (a maintained fork of proplot supporting mpl 3.9+). Every matplotlib axes method still works; UltraPlot adds a unified .format() styling method, smarter subplots(), location-shortcut colorbars/legends, constructor functions, and auto-layout.

Most of UltraPlot is close enough to matplotlib that you can guess it. This skill documents the conventions and edges that are easy to get subtly wrong — all verified against ultraplot 2.3.

Core conventions (get these right first)

• Import as uplt, not plt and not pplt: import ultraplot as uplt. (pplt is the old proplot alias; docs/old code may use it but uplt is current.)
• Don't use the pyplot interface. There is no uplt.plot() / uplt.gca() / uplt.show()-driven state machine. Always go through explicit figure/axes objects.
• Style with .format(), not set_* calls. One ax.format(...) / fig.format(...) call replaces dozens of set_xlabel/set_title/set_xlim/ tick calls. set_* methods still work, but .format() is the idiom.

import numpy as np
import ultraplot as uplt

fig, axs = uplt.subplots(nrows=2, ncols=2, share=True, refwidth=2.0)
fig.format(suptitle="Title", xlabel="x", ylabel="y", abc="a)", abcloc="ul")
axs[0].plot(np.random.rand(20, 3), cycle="538", labels=["a", "b", "c"], legend="ll")
m = axs[3].pcolormesh(np.random.rand(20, 20), cmap="magma")
fig.colorbar(m, loc="r", label="value")
fig.save("out.png", dpi=200)

Figure creation

• subplots() ALWAYS returns a SubplotGrid, even for one subplot. So fig, axs = uplt.subplots(ncols=1) makes axs a grid — index it as axs[0] to get the axes. Use the singular uplt.subplot() when you want one bare axes.
• Sizing is automatic. Prefer refwidth/refheight (per-subplot reference size) + refaspect; the figure size and inter-subplot spacing are computed by auto-layout. Use figwidth/figheight only to force the overall size.
• Units are flexible on every size arg: numbers in subplots/format default to inches for figure dims and font-relative em for subplot/spacing dims; strings accept 'cm', 'mm', 'in', 'pt', 'em' (e.g. refwidth='4cm', wspace='1em').

SubplotGrid indexing (common trap)

SubplotGrid is a list-like container. Indexing is not numpy-like:

To get a single axes object, use flat 1D integer indexing axs[i]. 2D and slice indexing return grids. Methods broadcast over a grid (axs.format(...), axs[0, :].plot(...) all work), so axs[i, j].plot() works — but axs[i, j] is a grid, so anything expecting a single Artist/Axes back will misbehave.

The .format() method

Works on axes, figures, and grids. Common keywords (all optional):

ax.format(
    title="t", ltitle=..., rtitle=..., ultitle=..., urtitle=...,  # corner titles
    xlabel="x", ylabel="y", xlim=(0, 10), ylim=..., xscale="log",
    xlocator="maxn", xformatter="sci",        # string shortcuts -> constructors
    xticks=[0, 5, 10], xticklabels=[...], xtickdir="inout", xticklen=5,
    xreverse=False, xmargin=0.05,
    grid=True, gridminor=True,
    abc="a)", abcloc="ul",                    # auto subplot labels: a) b) c) ...
)
fig.format(
    suptitle="figure title",
    leftlabels=["row1", "row2"], toplabels=["col1", "col2"],  # edge super-labels
    # ...also accepts every axes-format kwarg, applied to all subplots
)
axs.format(xlabel="shared", abc="A.")         # broadcasts to all axes in the grid

• abc=True (or a template like 'a)', 'A.', '(a)') turns on automatic subplot labels; abcloc placement uses corner codes 'ul' 'uc' 'ur' 'll' 'lc' 'lr'.
• xlocator/xformatter/xscale accept the same strings/tuples as the constructor functions below (e.g. xlocator=('maxn', 5)).

Colorbars & legends (location shortcuts)

ax.colorbar(...), fig.colorbar(...), ax.legend(...), fig.legend(...) all take a loc of single-letter / corner codes — not matplotlib's 'upper right' style (though those still work for legends):

m = ax.pcolormesh(z, cmap="viridis")
ax.colorbar(m, loc="r", label="v", length=0.8, width="1.5em")
fig.colorbar(m, loc="r", label="v")                 # figure-spanning colorbar

ax.legend(loc="ll", ncols=1)                        # from labeled artists
fig.legend(loc="b", ncols=3)

Easiest path: pass colorbar=/legend= (a loc code) straight to the plotting command and skip the separate call:

ax.pcolormesh(z, cmap="magma", colorbar="r")
ax.plot(y, labels=["a", "b", "c"], legend="ul")

Constructor functions

These turn strings/tuples/lists into matplotlib objects; the same inputs are accepted inline by plotting commands and .format().

uplt.Colormap("viridis", left=0.1, right=0.9, reverse=True)  # truncate/reverse/edit
uplt.Colormap(["red", "blue"])                               # build from colors
uplt.Cycle("538"); uplt.Cycle("viridis", 5); uplt.Cycle(["r","g","b"])
uplt.Norm("diverging", vcenter=0); uplt.Norm(("power", 2)); uplt.Norm("log")
uplt.Locator(("maxn", 5)); uplt.Locator("log")
uplt.Formatter("sci"); uplt.Formatter("frac"); uplt.Formatter("percent")
uplt.Scale(("power", 2)); uplt.Proj("ortho", central_latitude=45)

Colormaps, cycles, colors

Registered on import (case-insensitive; append _r to reverse). Verified names:

• Perceptually-uniform sequential: viridis, magma, plasma, inferno, cividis.
• Diverging: rdbu, coolwarm, piyg, brbg, spectral.
• Color cycles (pass to cycle=): default, colorblind, colorblind10, 538, ggplot, seaborn, bmh, tab10, tab20, Set1, Qual1, Qual2.

List what's available at runtime: uplt.show_cmaps(), uplt.show_cycles(), uplt.show_colors(), uplt.show_fonts(). Programmatic registry: uplt.colormaps.

Plotting commands (PlotAxes enhancements)

Standard matplotlib commands (plot, scatter, pcolormesh, contourf, ...) plus extras like ax.heatmap(...). Enhancements:

• Pass cmap=, cycle=, norm=, levels=, colorbar=, legend=, labels= directly to the plot call.
• Multi-column y → one line per column (with labels=[...] per column).
• pandas/xarray inputs auto-label axes/legend from names/coords.

Axis sharing & spanning, panels, insets

uplt.subplots(nrows=2, ncols=2,
    share=True,    # share=sharex=sharey level: 0/False .. 3/True, 4/'all', 'auto'
    span=True,     # single centered label spanning a row/column (spanx/spany)
    align=True)    # align labels across subplots (alignx/aligny)

ax.panel("r", width="3em")          # attached panel: 'l'/'r'/'t'/'b' (alias panel_axes)
ax.inset([0.5, 0.5, 0.4, 0.4])      # inset in axes-fraction coords (alias inset_axes)

Share levels: 0/False none · 1/'labels' labels only · 2/'limits' limits+ticks · 3/True + hide inner ticklabels · 4/'all' across all rows/cols · 'auto'.

Geographic & polar axes

fig, ax = uplt.subplots(proj="ortho", proj_kw=dict(central_latitude=45))  # needs cartopy
ax.format(land=True, coast=True, borders=True,
          lonlim=(-60, 60), latlim=(-30, 30), lonlocator=20, latlocator=20,
          longrid=True, latgrid=True)

fig, ax = uplt.subplots(proj="polar")   # PolarAxes, no cartopy needed
ax.format(thetalim=(0, 360), rlim=(0, 1), thetalocator=30, rlabelpos=45)

Geo projections ('ortho', 'merc', 'moll', 'robin', ...) require cartopy; without it subplots(proj=...) raises. proj="polar" always works.

rc configuration

uplt.rc["font.size"] = 11           # dict access (dotted keys)
uplt.rc.update({"axes.grid": True, "grid.alpha": 0.4})
with uplt.rc.context(fontsize=12, linewidth=1.5):  # underscore aliases; auto-reverts
    fig, ax = uplt.subplots()
uplt.rc.reset()

Loads defaults → ~/.config/ultraplotrc → ./.ultraplotrc (later overrides earlier).

Saving

fig.save("out.pdf", dpi=300) or fig.savefig(...) (alias). Default format pdf, default dpi 1000.

Common mistakes