Replace the logo and move the site onto the scikit-learn palette

README.md

@@ -1,11 +1,6 @@
-<!-- <script src="https://kit.fontawesome.com/d20edc211b.js" crossorigin="anonymous"></script>
+<img src="docs/img/logo.svg" alt="IKPyKit" width="120" align="left" style="margin-right: 20px;">
 
-<div style="margin-bottom: 10px;">
-    <img src="img/ikpykit_logo_1.jpg#only-light" align="left" style="margin-bottom: 20px; margin-top: 0px;">
-    <img src="img/ikpykit_logo_1.jpg#only-dark" align="left" style="margin-bottom: 20px; margin-top: 0px;">
-</div> -->
-
-<!-- <div style="clear: both;"></div> -->
+<div style="clear: both;"></div>
 
 ![Python](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13%20%7C%203.14-blue)
 [![PyPI](https://img.shields.io/pypi/v/ikpykit)](https://pypi.org/project/ikpykit/)

docs/README.md

@@ -3,14 +3,9 @@ description: >-
   IKPyKit is a scikit-learn compatible Python library implementing Isolation Kernel methods for anomaly detection, clustering and change detection.
 ---
 
-<!-- <script src="https://kit.fontawesome.com/d20edc211b.js" crossorigin="anonymous"></script>
+<img src="img/logo.svg" alt="IKPyKit" width="120" align="left" style="margin-right: 20px;">
 
-<div style="margin-bottom: 10px;">
-    <img src="img/ikpykit_logo_1.jpg#only-light" align="left" style="margin-bottom: 20px; margin-top: 0px;">
-    <img src="img/ikpykit_logo_1.jpg#only-dark" align="left" style="margin-bottom: 20px; margin-top: 0px;">
-</div> -->
-
-<!-- <div style="clear: both;"></div> -->
+<div style="clear: both;"></div>
 
 ![Python](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13%20%7C%203.14-blue)
 [![PyPI](https://img.shields.io/pypi/v/ikpykit)](https://pypi.org/project/ikpykit/)

docs/stylesheets/extra.css

@@ -208,3 +208,44 @@ Page Actions (Copy as Markdown)
   height: 0.9rem;
   fill: currentcolor;
 }
+
+/*
+===============================
+Brand colours
+===============================
+
+The palette in mkdocs.yml is set to "custom", which defers to the variables
+below. These are scikit-learn's blue and orange: IKPyKit implements the
+scikit-learn API, and that pair is how the surrounding ecosystem signals it.
+The logo is built from the same two values.
+*/
+
+:root {
+  --ik-blue: #3499cd;
+  --ik-blue-dark: #2a7fae;
+  --ik-blue-light: #6bb8e4;
+  --ik-orange: #f89939;
+  --ik-orange-dark: #e07f1f;
+}
+
+[data-md-color-scheme="default"] {
+  --md-primary-fg-color: var(--ik-blue);
+  --md-primary-fg-color--light: var(--ik-blue-light);
+  --md-primary-fg-color--dark: var(--ik-blue-dark);
+  --md-primary-bg-color: #ffffff;
+  --md-primary-bg-color--light: rgba(255, 255, 255, 0.7);
+  --md-accent-fg-color: var(--ik-orange-dark);
+  --md-accent-fg-color--transparent: rgba(248, 153, 57, 0.1);
+}
+
+[data-md-color-scheme="slate"] {
+  /* On the dark scheme the header goes near-black, so the blue is kept for
+     links and highlights only -- a mid blue header washes out against it. */
+  --md-primary-fg-color: #1b2027;
+  --md-primary-fg-color--light: var(--ik-blue-light);
+  --md-primary-fg-color--dark: #12161b;
+  --md-primary-bg-color: #ffffff;
+  --md-accent-fg-color: var(--ik-orange);
+  --md-accent-fg-color--transparent: rgba(248, 153, 57, 0.1);
+  --md-typeset-a-color: var(--ik-blue-light);
+}

mkdocs.yml

@@ -108,7 +108,7 @@ plugins:
   # to include the version prefix during a versioned deploy.
   - social:
       cards_layout_options:
-        background_color: "#3f51b5" # matches the indigo primary
+        background_color: "#3499cd" # matches the primary and the logo
   # Must come after mkdocstrings and mkdocs-jupyter: this plugin converts the
   # rendered HTML back to Markdown, so it needs their output to already exist.
   # Otherwise the API pages would export as bare `::: ikpykit.X` directives.
@@ -179,7 +179,7 @@ plugins:
 theme:
   name: material
   custom_dir: docs/overrides
-  favicon: img/favicon.png
+  favicon: img/logo.svg
   icon:
     repo: fontawesome/brands/github
   features:
@@ -198,17 +198,19 @@ theme:
     text: Open Sans
     code: Ubuntu Mono
   highlightjs: true
-  logo: img/favicon.png
+  logo: img/logo.svg
   palette:
     - scheme: default
-      primary: indigo
-      accent: indigo
+      # "custom" defers to the values in stylesheets/extra.css, which carry
+      # scikit-learn's exact blue and orange -- the same pair the logo uses.
+      primary: custom
+      accent: custom
       toggle:
         icon: material/weather-sunny
         name: Switch to dark mode
     - scheme: slate
-      primary: black
-      accent: black
+      primary: custom
+      accent: custom
       toggle:
         icon: material/weather-night
         name: Switch to light mode

scripts/gen_logo.py

@@ -0,0 +1,217 @@
+"""Draw the IKPyKit logo by running Isolation Kernel on a small point set.
+
+The mark is an "ik" monogram over a Voronoi partition, and the partition is not
+decoration: it is the output of one aNNE estimator, the same partition the
+kernel uses to measure similarity. Cells are shaded by area, and the largest --
+the one whose point sits furthest from any neighbour, and so is easiest to
+isolate -- is the orange one.
+
+The palette is scikit-learn's blue and orange. IKPyKit implements the
+scikit-learn API, and that pair is how the surrounding ecosystem signals it;
+the form is a monogram rather than scikit-learn's blobs, so the mark reads as a
+sibling rather than as something official.
+
+Nothing is hand-drawn, so the mark regenerates at any size and the palette or
+the cell count can be changed by editing the constants below.
+
+    python scripts/gen_logo.py            # write the assets into docs/img/
+    python scripts/gen_logo.py --preview  # contact sheet of other candidates
+"""
+
+from __future__ import annotations
+
+import argparse
+import pathlib
+
+import cairosvg
+import numpy as np
+from scipy.spatial import Voronoi
+
+from ikpykit.kernel import IsoKernel
+
+OUT_DIR = pathlib.Path("docs/img")
+SIZE = 512
+
+# scikit-learn's blue, lightened into a four step ramp. Their blue is a mid
+# tone, so the ramp deliberately stops short of navy -- a dark step reads as a
+# different family and weighs one corner of the mark down.
+BLUE = ("#9ed2ee", "#6bb8e4", "#3499cd", "#2a7fae")
+ORANGE = "#f89939"
+WHITE = "#ffffff"
+
+# Chosen from the candidates --preview renders, on composition: balanced cells,
+# no slivers, and the orange cell clear of the monogram's counters.
+CELLS = 4
+RANDOM_STATE = 3
+
+# The monogram, drawn as constant weight strokes. The k's arms are one path
+# whose vertex sits on the stem's centre line, and the stem is painted after
+# them, so the join between arms is hidden rather than spiking out to the left.
+STROKE = 52
+ARMS = "M 401 178 L 245 290 L 401 392"
+K_STEM = "M 245 120 L 245 392"
+I_STEM = "M 139 200 L 139 392"
+DOT = (139, 150, 31)
+
+
+def sample_points(random_state: int, n: int = 160) -> np.ndarray:
+    rng = np.random.RandomState(random_state)
+    centres = rng.uniform(0.2, 0.8, size=(4, 2))
+    points = np.vstack([c + rng.randn(n // 4, 2) * 0.13 for c in centres])
+    return np.clip(points, 0.06, 0.94)
+
+
+def partition(n_cells: int, random_state: int) -> list[np.ndarray]:
+    """Return the Voronoi cells one aNNE estimator lays over the unit square.
+
+    Mirroring the centres across each edge makes every cell finite and exactly
+    bounded by the square, which avoids clipping infinite ridges by hand.
+    """
+    centres = (
+        IsoKernel(
+            method="anne",
+            n_estimators=1,
+            max_samples=n_cells,
+            random_state=random_state,
+        )
+        .fit(sample_points(random_state))
+        .iso_kernel_.center_data
+    )
+    blocks = [centres]
+    for axis, value in ((0, 0.0), (0, 1.0), (1, 0.0), (1, 1.0)):
+        mirror = centres.copy()
+        mirror[:, axis] = 2 * value - mirror[:, axis]
+        blocks.append(mirror)
+    diagram = Voronoi(np.vstack(blocks))
+    return [
+        diagram.vertices[diagram.regions[diagram.point_region[i]]]
+        for i in range(len(centres))
+    ]
+
+
+def area(poly: np.ndarray) -> float:
+    x, y = poly[:, 0], poly[:, 1]
+    return 0.5 * abs(np.dot(x, np.roll(y, -1)) - np.dot(y, np.roll(x, -1)))
+
+
+def fatness(poly: np.ndarray) -> float:
+    """1 for a circle, near 0 for a sliver. Keeps thin wedges out of the mark."""
+    perimeter = np.hypot(*(np.roll(poly, -1, 0) - poly).T).sum()
+    return 4 * np.pi * area(poly) / perimeter**2 if perimeter else 0.0
+
+
+def composition_score(cells: list[np.ndarray]) -> float:
+    """Rank candidates so --preview shows the ones worth looking at first."""
+    areas = np.array([area(c) for c in cells])
+    return min(fatness(c) for c in cells) * 2 + (
+        1 - abs(areas.max() / areas.sum() - 0.4) * 4
+    )
+
+
+def render(
+    n_cells: int = CELLS, random_state: int = RANDOM_STATE, size: int = SIZE
+) -> str:
+    cells = partition(n_cells, random_state)
+    areas = np.array([area(c) for c in cells])
+    rank = np.argsort(np.argsort(-areas))
+    isolated = int(np.argmax(areas))
+
+    polygons = []
+    for i, cell in enumerate(cells):
+        points = " ".join(f"{x * size:.1f},{(1 - y) * size:.1f}" for x, y in cell)
+        fill = ORANGE if i == isolated else BLUE[min(rank[i], len(BLUE) - 1)]
+        polygons.append(f'<polygon points="{points}" fill="{fill}"/>')
+
+    cx, cy, r = DOT
+    joined = "\n        ".join(polygons)
+    return f"""<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {size} {size}" \
+role="img" aria-label="IKPyKit">
+  <defs>
+    <clipPath id="disc"><circle cx="256" cy="256" r="230"/></clipPath>
+  </defs>
+  <g id="partition" clip-path="url(#disc)">
+    <g stroke="{WHITE}" stroke-width="8" stroke-linejoin="round">
+        {joined}
+    </g>
+  </g>
+  <!-- The mark is mostly blue and the site header is too, so without this ring
+       the disc dissolves into it. On a light background the ring is invisible. -->
+  <circle cx="256" cy="256" r="230" fill="none" stroke="{WHITE}" stroke-width="16"/>
+  <g id="monogram">
+    <g stroke="{WHITE}" stroke-width="{STROKE}" fill="none" stroke-linejoin="round">
+      <path d="{ARMS}"/>
+      <path d="{K_STEM}"/>
+      <path d="{I_STEM}"/>
+    </g>
+    <!-- The ring keeps the dot visible when it falls on the orange cell. -->
+    <circle cx="{cx}" cy="{cy}" r="{r}" fill="{ORANGE}" stroke="{WHITE}" stroke-width="11"/>
+  </g>
+</svg>
+"""
+
+
+def write_assets() -> None:
+    OUT_DIR.mkdir(parents=True, exist_ok=True)
+    mark = render()
+    written = ["logo.svg"]
+    (OUT_DIR / "logo.svg").write_text(mark)
+    # PNG fallbacks: 512 for social cards and PyPI, 180 for iOS home screens,
+    # 32 for browsers that still ask for a raster favicon.
+    for name, width in (
+        ("logo-512.png", 512),
+        ("logo-180.png", 180),
+        ("favicon-32.png", 32),
+    ):
+        cairosvg.svg2png(
+            bytestring=mark.encode(), write_to=str(OUT_DIR / name), output_width=width
+        )
+        written.append(name)
+    print("\n".join(f"wrote {OUT_DIR / name}" for name in written))
+
+
+def write_preview(n_cells: int, count: int) -> None:
+    import io
+
+    from PIL import Image
+
+    ranked = sorted(
+        ((composition_score(partition(n_cells, s)), s) for s in range(40)),
+        key=lambda pair: -pair[0],
+    )[:count]
+
+    sheet = Image.new("RGBA", (220 * count, 290), (255, 255, 255, 255))
+    for i, (score, state) in enumerate(ranked):
+        png = cairosvg.svg2png(
+            bytestring=render(n_cells, state).encode(), output_width=210
+        )
+        big = Image.open(io.BytesIO(png)).convert("RGBA")
+        flat = Image.new("RGBA", big.size, (255, 255, 255, 255))
+        flat.alpha_composite(big)
+        sheet.alpha_composite(flat, (220 * i, 0))
+        for j, px in enumerate((16, 24, 32)):
+            sheet.alpha_composite(
+                flat.resize((px, px), Image.LANCZOS), (220 * i + 10 + j * 60, 225)
+            )
+        print(f"random_state={state:<3} score={score:.3f}")
+    path = pathlib.Path(f"/tmp/ikpykit-logo-{n_cells}-cells.png")
+    sheet.save(path)
+    print(f"contact sheet: {path}")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--preview", action="store_true", help="render other candidates"
+    )
+    parser.add_argument("--cells", type=int, default=CELLS)
+    parser.add_argument("--count", type=int, default=6)
+    args = parser.parse_args()
+
+    if args.preview:
+        write_preview(args.cells, args.count)
+    else:
+        write_assets()
+
+
+if __name__ == "__main__":
+    main()