"""Reproduce the synthetic-data figure from the recorded production skill session. The coding agent creates this file locally; SkillGild validates the palette. Run beside synthetic_accuracy.csv with the versions in requirements.txt. """ import csv import json from collections import defaultdict from pathlib import Path import matplotlib matplotlib.use("Agg") import matplotlib.pyplot as plt import numpy as np ROOT = Path(__file__).resolve().parent COLORS = {"Baseline": "#6D6A8B", "Example method": "#2467BD"} matplotlib.rcParams.update( { "font.family": "DejaVu Serif", "font.size": 7.5, "text.color": "#2B2A3D", "axes.labelcolor": "#2B2A3D", "axes.edgecolor": "#5D5B78", "xtick.color": "#5D5B78", "ytick.color": "#5D5B78", "svg.fonttype": "none", "svg.hashsalt": "skillgild-production-plotting-example", "pdf.fonttype": 42, } ) runs = defaultdict(lambda: defaultdict(list)) with (ROOT / "synthetic_accuracy.csv").open() as file: for row in csv.DictReader(file): runs[row["method"]][int(row["step"])].append(float(row["val_accuracy_pct"])) fig, ax = plt.subplots(figsize=(3.25, 2.5)) fig.subplots_adjust(left=0.17, right=0.70, bottom=0.18, top=0.84) last_means = {} for method, by_step in runs.items(): steps = sorted(by_step) if any(len(by_step[step]) != 3 for step in steps): raise ValueError("Each method and checkpoint must contain exactly three seeds") xs = np.array(steps) / 1000 mean = np.array([np.mean(by_step[step]) for step in steps]) std = np.array([np.std(by_step[step], ddof=1) for step in steps]) color = COLORS[method] ax.fill_between(xs, mean - std, mean + std, color=color, alpha=0.12, lw=0) ax.plot( xs, mean, color=color, linestyle="--" if method == "Baseline" else "-", marker="s" if method == "Baseline" else "o", markersize=2.8, linewidth=1.1, markerfacecolor="white", markeredgewidth=0.9, ) ax.annotate( f"{'Example' if method == 'Example method' else method}\n{mean[-1]:.1f}%", (xs[-1], mean[-1]), xytext=(5, 6 if method == "Example method" else -7), textcoords="offset points", color=color, fontsize=7, va="center", annotation_clip=False, ) last_means[method] = float(mean[-1]) ax.set_xlim(0, 16) ax.set_ylim(0, 90) ax.set_xticks([0, 4, 8, 12, 16]) ax.set_xlabel("Training steps (thousands)", fontsize=7.5) ax.set_ylabel("Validation accuracy (%)", fontsize=7.5) ax.grid(axis="y", color="#E4E1F2", linewidth=0.4) ax.set_axisbelow(True) ax.spines[["top", "right"]].set_visible(False) fig.suptitle("Synthetic data: mean ± s.d., 3 seeds", x=0.04, ha="left", fontsize=7) fig.savefig(ROOT / "accuracy.svg", metadata={"Date": None}) fig.savefig(ROOT / "accuracy.pdf", metadata={"CreationDate": None, "ModDate": None}) fig.savefig(ROOT / "accuracy.png", dpi=300) plt.close(fig) print(json.dumps({"final_mean_accuracy_pct": last_means, "size_inches": [3.25, 2.5]}))