medusa-style

Single source of truth (SSOT) for MEDUSA styling. One look themes BOTH PySide6 desktop apps (via generated QSS) and Matplotlib scientific plots (built on Matplotlib's bundled seaborn baseline) — so a microvolt is the same color in the dark acquisition app and on a white-paper export, and an embedded plot is seamless with the panel that hosts it.

The look has two independent axes:

• a theme — the color scheme (dark, light, …), a Palette.
• a style — the visual language (modern, presentation, …): fonts, density and plot line widths, a Style.

Any style renders in any theme. The code is organized by rendering environment (qt/, mpl/, …), so adding a backend never touches the others.

Install

pip install git+https://github.com/medusabci/medusa-style.git

That's all — PySide6 and Matplotlib are both core dependencies (every MEDUSA component needs both). The only extra is [dev] for building/testing the package itself. The seaborn library is not required: the seaborn base styles ship inside Matplotlib.

The whole API

import medusa_style as ms

ms.apply(app)               # paint a Qt application (widgets + seamless plots)
ms.apply(widget)            # scope the theme to one Qt widget
ms.apply()                  # theme every Matplotlib figure you draw
ms.apply(fig)               # restyle one figure in place
ms.apply(ax)                # restyle one axes in place
ms.use_theme("light")       # switch COLORS — restyles everything already styled
ms.use_style("presentation")# switch VISUAL LANGUAGE — restyles everything
ms.current_theme()          # the active Palette
ms.current_style()          # the active Style
ms.themes()                 # ['dark', 'light']          — for a theme picker
ms.styles()                 # ['modern', 'presentation'] — for a style picker
ms.customize(colors=..., style=...)   # one-line house theme / style
ms.on_change(cb)            # react to switches — cb(palette, style)
ms.categorical_color(7)     # SSOT categorical color #7

apply() dispatches on what you give it: nothing → global Matplotlib, a QApplication/QWidget → Qt, a Figure/Axes → Matplotlib. You never choose a backend. Whatever you pass is remembered, so a later use_theme()/use_style() repaints it with no loop on your side.

Desktop app (PySide6)

import sys
import medusa_style as ms
from medusa_style.qt import application, PlotPanel
from PySide6.QtWidgets import QMainWindow

app = application(sys.argv)        # Hi-DPI handled before construction; themed
ms.apply()                         # theme Matplotlib figures too

win = QMainWindow()
panel = PlotPanel(toolbar=True)    # seamless embed — no objectName, no restyle call
panel.figure.add_subplot(111).plot(range(10), color=ms.categorical_color(0))
win.setCentralWidget(panel)
win.show()

# View menu -> app AND embedded figure restyle, one line each:
view = win.menuBar().addMenu("View")
view.addAction("Light", lambda: ms.use_theme("light"))
view.addAction("Presentation", lambda: ms.use_style("presentation"))
sys.exit(app.exec())

PlotPanel is the borderless Matplotlib↔Qt seam as a drop-in widget: its background equals the plot's, and it tracks itself for live theme/style switches. For a frame you already own, use medusa_style.qt.embed_figure(frame, fig).

Notebook / headless (Qt never imported)

import medusa_style as ms
ms.apply()                  # theme Matplotlib globally
fig = make_figure()
ms.apply(fig)               # ...or restyle one figure
ms.use_theme("dark")        # restyles tracked figures in place

Importing medusa_style pulls in neither PySide6 nor Matplotlib until you actually style something.

Themes and styles

Two built-in themes (color schemes), both tuned for long medical/scientific sessions:

• dark (default) — deep, cool, near-black canvas for dim-lab real-time use.
• light — a calm cool-gray elevation ramp for reports/paper export (plot_bg is pure white).

Two built-in styles (visual languages):

• modern (default) — the clean flat MEDUSA look (assets/qss/base.qss).
• presentation — larger type, thicker plot lines and roomier controls for projectors and slides (token-only; reuses base.qss).

The data-visualization encodings — a 16-color colorblind-aware categorical cycle and the medusa_sequential (PSD) / medusa_diverging (topomap) colormaps — are theme- and style-independent, so results stay comparable across every look.

Customize

Zero config by default; override in one line. colors derive a new theme, style derives a new style:

ms.customize(
    colors={"accent_primary": "#FF6600"},                # -> a registered theme
    style={"ui_size_body": 15, "plot_line_width": 1.4},  # -> a registered style
)                            # activates + live-restyles; returns the created value(s)

Architecture & extensibility

medusa_style/
  api.py        the public verbs
  dispatch.py   target detection + environment routing (stdlib-only)
  appearance.py the active (theme, style) + live switching + observer
  palette.py    the color SSOT (Palette, DARK, LIGHT) + encodings
  style.py      the visual languages (Style, MODERN, PRESENTATION)
  resources.py  zip-safe asset loading
  qt/           the Qt environment (QSS render, PlotPanel, icons, fonts)
  mpl/          the Matplotlib environment (rcParams, colormaps, figure styling)
  assets/       qss/ (structural templates), icons/, fonts/, brand/

• Add a theme — one Palette(...) literal in palette.py (+ PALETTES entry) or ms.register_theme(...). Works in every environment immediately.
• Add a style — one Style(...) literal in style.py (+ STYLES entry) or ms.register_style(...). A token-only style is pure data; only a structurally different widget language needs its own assets/qss/<name>.qss.
• Add an environment (Plotly, VisPy, Bokeh, …) — a new package with a matches/_apply/reapply trio and one ms.register_environment(...). No existing file changes.

Power-user namespaces

The tiny top-level surface is enough for almost everything; reach into these only when you need the knobs:

medusa_style.qt imports PySide6 (never Matplotlib); medusa_style.mpl imports Matplotlib (never Qt). import medusa_style loads neither until you style something, so a headless figure-export job never pulls in Qt.

Icons, fonts & brand assets

from medusa_style import qt

button.setIcon(qt.icon("save_as"))        # theme-agnostic action icon
button.setIcon(qt.icon("delete_forever", color="error"))  # ...or a semantic color
splash = qt.pixmap("medusa_splash")       # brand images (splash/login/about)
win.setWindowIcon(qt.app_icon())          # the MEDUSA app icon
path = qt.asset_path("fonts/Roboto-Regular.ttf")   # any bundled file, real path

qt.icon(name) icons are theme-agnostic. The color is not baked in: each icon recolors to the active theme on every repaint and changes with Qt's state automatically — a disabled widget dims its icon, and a theme switch repaints it live, with no re-fetch. Pass color= (a palette role like "error"/"accent_primary", or a "#RRGGBB" literal) to force a specific color; it still dims when disabled.

qt.application() does the common setup for you: it registers the bundled fonts (Roboto family, Abel, Dense, Womby) and sets the MEDUSA window icon app-wide (both opt-out via application(set_icon=False, load_bundled_fonts=False)).

Packaging

All runtime data lives under medusa_style/assets/ — qss/ (the structural QSS templates), icons/, fonts/ and brand/. They are declared in pyproject.toml so the wheel always includes them, while the editable .psd design sources stay in the repo-level design_source/ directory (outside the package, never shipped):

[tool.hatch.build.targets.wheel]
packages  = ["medusa_style"]
artifacts = [
    "medusa_style/assets/**/*.qss",
    "medusa_style/assets/**/*.svg",
    "medusa_style/assets/**/*.png",
    "medusa_style/assets/**/*.ico",
    "medusa_style/assets/**/*.ttf",
    "medusa_style/assets/**/*.otf",
    "medusa_style/py.typed",
]

At runtime the assets are loaded zip-safely via medusa_style.resources (which wraps importlib.resources), so it works even from inside a zipped wheel. Matplotlib .mplstyle files are generated on demand (medusa_style.mpl.write_mplstyle(...)) rather than shipped, so they can never drift from the SSOT.