data-visualization

name	data_visualization
description	Use this skill whenever the task involves making a figure. Produces publication-quality matplotlib charts (bar, scatter, line) with a colorblind-safe palette, saves them, and self-checks the result with `figure_check`.

Data Visualization

Every figure should be clear, colorblind-safe, and self-contained (readable without the surrounding text). This skill gives you a small toolkit to reach that bar without writing matplotlib boilerplate from scratch.

The recommended workflow is:

1. Apply the publication theme once per session.
2. Pick a recipe (bar / scatter / line) and adapt it.
3. Save as PNG (and SVG if you want a vector copy).
4. Run figure_check to get a vision-model review — fix any issues before showing the user.

---

1. Global theme — apply FIRST in every session

import matplotlib
import matplotlib.pyplot as plt

def set_publication_theme(base_font_size: int = 8, line_width: float = 0.75):
    """Apply a clean, publication-quality matplotlib theme."""
    matplotlib.rcParams.update({
        # Fonts
        "font.family": "sans-serif",
        "font.sans-serif": ["Arial", "Helvetica", "DejaVu Sans"],
        "font.size": base_font_size,
        "axes.titlesize": base_font_size + 1,
        "axes.labelsize": base_font_size,
        "xtick.labelsize": base_font_size - 1,
        "ytick.labelsize": base_font_size - 1,
        "legend.fontsize": base_font_size - 1,
        # Axes
        "axes.linewidth": line_width,
        "axes.spines.top": False,
        "axes.spines.right": False,
        "axes.grid": False,
        # Ticks
        "xtick.major.width": line_width,
        "ytick.major.width": line_width,
        "xtick.direction": "out",
        "ytick.direction": "out",
        # Save
        "figure.dpi": 150,
        "savefig.dpi": 300,
        "savefig.bbox": "tight",
        # Editable text in PDF/SVG (Illustrator/Inkscape)
        "pdf.fonttype": 42,
        "svg.fonttype": "none",
        # Legend
        "legend.frameon": False,
    })

set_publication_theme()

Key rules baked into the theme:

• No top / right spines.
• No legend frame.
• Editable text when exported to PDF or SVG.

---

2. Colorblind-safe palette (Wong, 2011)

PALETTE = [
    "#0072B2",  # blue
    "#E69F00",  # orange
    "#009E73",  # green
    "#CC79A7",  # pink
    "#56B4E9",  # sky blue
    "#D55E00",  # vermilion
    "#F0E442",  # yellow
    "#000000",  # black
]

Never use matplotlib's default colour cycle or jet. For >8 categories, encode shape + colour together.

---

3. Recipes

Bar chart (comparing values across a few items)

fig, ax = plt.subplots(figsize=(4.5, 3))
ax.bar(labels, values, color=PALETTE[:len(values)], width=0.6, linewidth=0)
ax.set_ylabel("Molecular weight (g/mol)")
ax.set_title("Comparison of drug MWs")
ax.tick_params(axis="x", rotation=20)
fig.tight_layout()
fig.savefig("mw_comparison.png", dpi=300)
plt.close(fig)

Scatter (two numeric variables)

fig, ax = plt.subplots(figsize=(4, 3))
ax.scatter(x, y, s=20, alpha=0.75, color=PALETTE[0], linewidths=0)
ax.set_xlabel("LogP")
ax.set_ylabel("Molecular weight (g/mol)")
fig.tight_layout()
fig.savefig("scatter.png", dpi=300)
plt.close(fig)

Line (trend / dose-response)

fig, ax = plt.subplots(figsize=(4.5, 3))
for i, (label, y) in enumerate(series.items()):
    ax.plot(x, y, color=PALETTE[i], marker="o", markersize=4, linewidth=1.2, label=label)
ax.set_xlabel("Dose (mg)")
ax.set_ylabel("Response (%)")
ax.legend(title="Compound")
fig.tight_layout()
fig.savefig("dose_response.png", dpi=300)
plt.close(fig)

---

4. Saving — always close the figure

fig.savefig("plot.png", dpi=300)   # raster, good default for notebooks
fig.savefig("plot.svg")            # vector, editable
plt.close(fig)                     # release memory

After saving, tell the user the file path — that is the deliverable.

---

5. Self-check (MANDATORY)

Always review the saved figure with figure_check and fix any issues it flags:

from utils.codeact_skills.skills.data_visualization.scripts.figure_check import figure_check

feedback = figure_check("mw_comparison.png")
for dimension, comment in feedback.items():
    print(f"{dimension}: {comment}")

figure_check returns a dict with five short comments: Readability, Panel Arrangement, Axis Labels, Legend, Color. If anything looks wrong, regenerate the figure with the fix applied before reporting back to the user.

---

6. Common mistakes to avoid

❌	✅
Default matplotlib colors	Wong 2011 palette
frameon=True on legend	frameon=False (already in the theme)
Top/right spines visible	Removed by the theme
fontsize=12 for small panels	7–8 pt body, 9 pt panel labels
Axis label without units	"Molecular weight (g/mol)"
Forgetting plt.close(fig)	Always close after saving
Skipping figure_check	Always run it before reporting