| """Render a composition graph for a Wordmaze config, for the dataset card. |
| |
| uv run plot_dataset_stats.py [data/wordmaze-m4-6/metadata.json] [-o dataset_stats.png] |
| |
| Reads the metadata.json a generation run writes and produces two panels: puzzle counts |
| across (word length x move count), and the vocabulary -> ladder-graph sizes per length. |
| """ |
| |
| |
| |
| |
| from __future__ import annotations |
|
|
| import argparse |
| import json |
| from pathlib import Path |
|
|
| import matplotlib.pyplot as plt |
| import numpy as np |
| from matplotlib import rcParams |
|
|
| INDIGO = "#6366f1" |
| INDIGO_DK = "#4f46e5" |
| INDIGO_LT = "#a5a6f6" |
| GRAY = "#9ca3af" |
| INK = "#1a1a1a" |
|
|
|
|
| def main() -> None: |
| here = Path(__file__).resolve().parent |
| ap = argparse.ArgumentParser(description="Plot Wordmaze dataset composition.") |
| ap.add_argument("metadata", nargs="?", default=str(here / "data" / "wordmaze-m4-6" / "metadata.json")) |
| ap.add_argument("-o", "--out", default=None, help="Output PNG (default: next to metadata.json).") |
| args = ap.parse_args() |
|
|
| meta = json.loads(Path(args.metadata).read_text()) |
| out = Path(args.out) if args.out else Path(args.metadata).with_name("dataset_stats.png") |
| slug = meta.get("move_count_slug", "wordmaze") |
|
|
| |
| combo = meta.get("examples_by_combo", {}) |
| lengths = sorted({int(k.split("_")[0]) for k in combo}) |
| moves = sorted({int(k.split("/")[1].split("_")[0]) for k in combo}) |
| wbl = {int(k): v for k, v in meta.get("words_by_length", {}).items()} |
| nodes = {int(k): v for k, v in meta.get("graph_nodes_by_length", {}).items()} |
| splits = meta.get("splits", {}) |
| total = meta.get("generated_examples", sum(combo.values())) |
|
|
| rcParams.update({ |
| "font.family": "DejaVu Sans", "font.size": 9.5, |
| "axes.spines.top": False, "axes.spines.right": False, |
| "axes.grid": True, "axes.axisbelow": True, "grid.color": "#ececec", |
| "axes.facecolor": "white", "figure.facecolor": "white", |
| "xtick.color": "#6b7280", "ytick.color": "#6b7280", |
| "xtick.major.size": 0, "ytick.major.size": 0, |
| }) |
| shades = {lengths[i]: [INDIGO_LT, INDIGO, INDIGO_DK][i % 3] for i in range(len(lengths))} |
|
|
| fig, (axA, axB) = plt.subplots(1, 2, figsize=(11, 4.6)) |
|
|
| |
| x = np.arange(len(moves)) |
| w = 0.8 / max(1, len(lengths)) |
| for i, L in enumerate(lengths): |
| vals = [combo.get(f"{L}_letters/{m}_moves", 0) for m in moves] |
| axA.bar(x + (i - (len(lengths) - 1) / 2) * w, vals, w, color=shades[L], |
| edgecolor="white", label=f"{L} letters", zorder=3) |
| axA.set_xticks(x) |
| axA.set_xticklabels([f"{m} moves" for m in moves]) |
| axA.set_ylabel("puzzles") |
| split_txt = " · ".join(f"{k} {v}" for k, v in sorted(splits.items())) |
| axA.set_title(f"{total:,} puzzles, balanced across length x moves", fontsize=11, fontweight="bold", color=INK, pad=10) |
| axA.legend(fontsize=8.5, loc="lower center", ncol=len(lengths), frameon=False) |
| axA.set_ylim(0, max(combo.values()) * 1.25 if combo else 1) |
|
|
| |
| xb = np.arange(len(lengths)) |
| axB.bar(xb - 0.2, [wbl.get(L, 0) for L in lengths], 0.4, color=GRAY, edgecolor="white", label="words in dictionary", zorder=3) |
| axB.bar(xb + 0.2, [nodes.get(L, 0) for L in lengths], 0.4, color=INDIGO, edgecolor="white", label="words with a neighbor", zorder=3) |
| axB.set_xticks(xb) |
| axB.set_xticklabels([f"{L} letters" for L in lengths]) |
| axB.set_ylabel("count") |
| axB.set_title("Vocabulary → ladder graph", fontsize=11, fontweight="bold", color=INK, pad=10) |
| axB.legend(fontsize=8.5, loc="upper right", frameon=False) |
|
|
| fig.suptitle(f"Wordmaze {slug} · wordfreq top-10k, proper nouns filtered", |
| fontsize=12.5, fontweight="bold", color=INK, y=1.02) |
| out.parent.mkdir(parents=True, exist_ok=True) |
| fig.savefig(out, dpi=200, bbox_inches="tight", facecolor="white") |
| print(f"saved {out}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
