Python Visualization

Routing Table

Load only the references needed for the task. Multiple files often apply.

Matplotlib (core)

Seaborn (statistical)

Publication Quality

Plotly (interactive)

Animations

Pandas styled tables

Biomedical visualization

LaTeX integration

Scripts

Style Assets

Pre-configured matplotlib style files in assets/:

Apply with plt.style.use('assets/nature.mplstyle') or via scripts/style_presets.py.

Decision Guide

Which library for the task?

Always prefer the matplotlib Object-Oriented interface (fig, ax = plt.subplots()) over the pyplot stateful interface for anything beyond a throwaway one-liner.

Seaborn builds on matplotlib. You can always drop down to matplotlib for fine-tuning after creating a seaborn plot — they share the same Figure/Axes objects.

Label Placement Rules

LLM-generated matplotlib code has a systematic blind spot: labels are generated without visual feedback, so small errors accumulate and produce collisions that are invisible in code but immediately obvious in the rendered figure. Follow these rules for every chart with text annotations.

Offsets must be relative, not hardcoded

Never use hardcoded data-unit offsets for annotations. Hardcoded values work for example data and break silently when units or scale change.

# BAD — breaks when y-axis is 0-1 or 0-10000
ax.annotate('IC50', xy=(ic50, mid_y), xytext=(ic50*5, mid_y + 15))

# GOOD — stays on-canvas for any scale
ylo, yhi = ax.get_ylim()
y_offset = 0.15 * (yhi - ylo)
ax.annotate('IC50', xy=(ic50, mid_y), xytext=(ic50*5, mid_y + y_offset))

# EVEN BETTER — use axes fraction for text, data coords for the arrow target
ax.annotate('Cmax', xy=(tmax, cmax), xytext=(0.6, 0.85),
            xycoords='data', textcoords='axes fraction',
            arrowprops=dict(arrowstyle='->'))

Stagger labels at the same x/y position

When multiple vertical or horizontal lines share the same label y/x position, stagger them:

# BAD — all three at identical y → collision when lines are close
for val, lbl in [(d10,'D10'), (d50,'D50'), (d90,'D90')]:
    ax.text(val, ax.get_ylim()[1]*0.95, lbl, ha='center')

# GOOD — alternating lanes
ylo, yhi = ax.get_ylim()
y_levels = [0.92, 0.76, 0.60]
for (val, lbl), y_frac in zip([(d10,'D10'),(d50,'D50'),(d90,'D90')], y_levels):
    ax.text(val, ylo + y_frac*(yhi - ylo), lbl, ha='center',
            bbox=dict(boxstyle='round,pad=0.1', fc='white', alpha=0.6, lw=0))

Gate per-cell annotations on matrix dimensions

Dense heatmaps with text annotations on every cell become illegible fast. Only annotate if both dimensions are ≤ 15:

rows, cols = matrix.shape
if rows <= 15 and cols <= 15:
    fontsize = max(5, 9 - max(rows, cols) // 3)
    for i in range(rows):
        for j in range(cols):
            ax.text(j, i, f'{matrix[i,j]:.2f}', ha='center', va='center',
                    fontsize=fontsize)

Significance brackets: derive position from data

yhi = ax.get_ylim()[1]
bar_top = bar_height + error
y = bar_top + 0.04 * yhi    # 4% breathing room
h = 0.02 * yhi              # bracket arm height
ax.plot([x1, x1, x2, x2], [y, y+h, y+h, y], 'k-', lw=0.8)
ax.text((x1+x2)/2, y+h, '***', ha='center', va='bottom', fontsize=7)

Panel labels: know when to increase x-offset

Standard offset -0.15 in axes fraction works for short y-labels. If your y-axis label is long (e.g., "Plasma Concentration (ng/mL)"), use -0.25 or more to avoid overlap:

ax.text(-0.15, 1.05, 'A', transform=ax.transAxes, fontsize=10, fontweight='bold')
# With long y-label: use -0.25 or -0.30

Core Patterns

Standard matplotlib figure

import matplotlib.pyplot as plt
import numpy as np

fig, ax = plt.subplots(figsize=(10, 6))
x = np.linspace(0, 2*np.pi, 100)
ax.plot(x, np.sin(x), label='sin(x)')
ax.set_xlabel('x')
ax.set_ylabel('y')
ax.legend(frameon=False)
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
fig.savefig('figure.png', dpi=300, bbox_inches='tight')

Seaborn statistical plot

import seaborn as sns
import matplotlib.pyplot as plt

df = sns.load_dataset('tips')
sns.scatterplot(data=df, x='total_bill', y='tip', hue='day')
plt.show()

Publication-ready figure

import matplotlib.pyplot as plt
plt.style.use('assets/nature.mplstyle')

fig, ax = plt.subplots(figsize=(3.5, 2.5))  # Nature single-column width
# ... plotting code ...
from figure_export import save_publication_figure
save_publication_figure(fig, 'figure1', formats=['pdf', 'png'], dpi=300)

Interactive Plotly chart

import plotly.express as px

df = px.data.gapminder().query("year == 2007")
fig = px.scatter(df, x="gdpPercap", y="lifeExp", size="pop",
                 color="continent", hover_name="country", log_x=True)
fig.show()
fig.write_html("figure.html")

Animated matplotlib → GIF

import matplotlib.pyplot as plt
import matplotlib.animation as animation
import numpy as np

fig, ax = plt.subplots()
x = np.linspace(0, 2*np.pi, 100)
line, = ax.plot([], [])
ax.set_xlim(0, 2*np.pi); ax.set_ylim(-1, 1)

def animate(i):
    line.set_data(x[:i], np.sin(x[:i]))
    return line,

ani = animation.FuncAnimation(fig, animate, frames=len(x), interval=30, blit=True)
ani.save('sine.gif', writer='pillow', fps=30)

Styled pandas table

import pandas as pd

styled = (df.style
    .format(precision=2)
    .highlight_max(color='lightgreen')
    .highlight_min(color='lightcoral')
    .background_gradient(subset=['value'], cmap='Blues'))
styled.to_html('table.html')

RDKit molecular grid

from rdkit import Chem
from rdkit.Chem import Draw

smiles = ['CCO', 'c1ccccc1', 'CC(=O)O']
mols = [Chem.MolFromSmiles(s) for s in smiles]
img = Draw.MolsToGridImage(mols, molsPerRow=3, subImgSize=(300, 300))
img.save('molecules.png')

Feedback Loop

This skill uses a feedback log to improve over time. The cycle:

1. Detect — After completing a task, note anything suboptimal.
2. Search — Check FEEDBACK.md for prior entries on the same topic.
3. Scope — Keep entries to 1–3 lines: category tag, what happened, what should change.
4. Draft & Ask — Draft the entry and show it to the user before writing.
5. Write on Approval — Append to FEEDBACK.md.
6. Compact at 75 — Summarize older entries when the file reaches 75 entries.

Always read FEEDBACK.md when reading this skill.