Matplotlib

matplotlib Best Practices

Matplotlib is the bedrock of Python data visualization. Adhering to these guidelines ensures your plots are not just visually appealing, but also robust, performant, and easily integrated into production AI/ML and data science pipelines.

1. Code Organization & Structure

1.1 Standard Imports

Always use the conventional aliases. This improves readability and consistency across projects.

❌ BAD:

import numpy
import matplotlib.pyplot
import matplotlib

✅ GOOD:

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

1.2 Object-Oriented API First

Prioritize the object-oriented API (Figure and Axes objects) over pyplot's stateful interface. This leads to more explicit, reproducible, and testable code, especially in functions or classes.

❌ BAD (Stateful pyplot):

plt.plot([1, 2, 3], [4, 5, 6])
plt.title("My Plot")
plt.xlabel("X-axis")
plt.ylabel("Y-axis")
plt.show() # Blocks execution

✅ GOOD (Object-Oriented):

from typing import List
import matplotlib.figure as mpl_figure
import matplotlib.axes as mpl_axes

def create_my_plot(data_x: List[float], data_y: List[float], title: str, x_label: str, y_label: str) -> mpl_figure.Figure:
    fig, ax = plt.subplots(figsize=(8, 6))
    ax.plot(data_x, data_y)
    ax.set_title(title)
    ax.set_xlabel(x_label)
    ax.set_ylabel(y_label)
    return fig

# Usage in a script or notebook
if __name__ == "__main__":
    x = [1, 2, 3]
    y = [4, 5, 6]
    my_figure = create_my_plot(x, y, "My Plot", "X-axis", "Y-axis")
    my_figure.savefig("my_plot.png") # Non-blocking save
    # If interactive display is needed, use:
    # plt.show()

1.3 Consistent Variable Naming

Use the standard variable names for Figure and Axes objects.

❌ BAD:

my_figure_object, my_axis_object = plt.subplots()

✅ GOOD:

fig, ax = plt.subplots()
# For multiple axes
fig, axs = plt.subplots(2, 2)

1.4 rcParams Access

Always access rcParams via mpl.rcParams to avoid import issues in early-loaded modules.

❌ BAD:

from matplotlib import rcParams
rcParams['figure.figsize'] = (10, 6)

✅ GOOD:

import matplotlib as mpl
mpl.rcParams['figure.figsize'] = (10, 6)

2. Common Patterns & Anti-patterns

2.1 Handling Dense Data

For dense scatter plots, use alpha for transparency. For very large collections, rasterize them to keep file sizes manageable.

✅ GOOD:

# Dense scatter plot with transparency
ax.scatter(X, Y, s=40, c='C1', alpha=0.1)

# Rasterizing large collections for PDF/SVG output
ax.scatter(X_large, Y_large, rasterized=True)
fig.savefig("rasterized_figure.pdf", dpi=600)

2.2 Publication-Quality Styling

Use custom linestyles with rounded caps and text path effects for professional figures.

✅ GOOD:

# Rounded dotted lines
ax.plot([0, 1], [0, 0], "C1", linestyle=(0, (0.01, 1)), dash_capstyle="round")

# Text outline for visibility
import matplotlib.patheffects as fx
text = ax.text(0.5, 0.1, "Label", transform=ax.transAxes)
text.set_path_effects([
    fx.Stroke(linewidth=3, foreground='white'),
    fx.Normal()
])

2.3 Managing Plot Margins

Always call fig.tight_layout() before saving or showing a figure to automatically adjust subplot parameters for a tight layout.

❌ BAD:

fig, ax = plt.subplots()
ax.plot(x, y)
fig.savefig("plot_with_margins.png") # May have excessive whitespace

✅ GOOD:

fig, ax = plt.subplots()
ax.plot(x, y)
fig.tight_layout() # Adjusts subplot params for tight layout
fig.savefig("plot_tight.png")

2.4 **kwargs Usage

Reserve **kwargs for pass-through arguments. Explicitly define keyword-only arguments for parameters consumed locally.

❌ BAD:

def my_plot_function(x, y, **kwargs):
    label = kwargs.pop('label', 'default') # Consuming from kwargs
    ax.plot(x, y, label=label, **kwargs)

✅ GOOD:

def my_plot_function(x, y, *, label='default', **kwargs): # Keyword-only 'label'
    ax.plot(x, y, label=label, **kwargs)

3. Performance Considerations

3.1 Avoid plt.show() in Libraries/Functions

plt.show() blocks execution and is intended for interactive display. In reusable code, return the Figure object or save it directly. This allows callers to decide how to handle the figure.

❌ BAD:

def generate_plot(data):
    fig, ax = plt.subplots()
    ax.plot(data)
    plt.show() # Blocks execution, not suitable for libraries

✅ GOOD:

def generate_plot(data) -> mpl_figure.Figure:
    fig, ax = plt.subplots()
    ax.plot(data)
    return fig

# In main script:
if __name__ == "__main__":
    my_data = [1, 5, 2, 8]
    plot_fig = generate_plot(my_data)
    plot_fig.savefig("my_generated_plot.pdf")
    # If interactive display is desired:
    # plt.show()

3.2 Explicitly Close Figures

When creating many figures programmatically (e.g., in loops or batch processing), explicitly close them to prevent memory leaks.

✅ GOOD:

for i in range(100):
    fig, ax = plt.subplots()
    ax.plot([0, i], [i, 0])
    fig.savefig(f"plot_{i}.png")
    plt.close(fig) # Essential for memory management

4. Type Hints

Embrace type hints for all new and modified public APIs. Use stub files (.pyi) where appropriate, or inline for simpler modules like pyplot.py.

✅ GOOD:

from typing import List
import matplotlib.figure as mpl_figure
import matplotlib.axes as mpl_axes

def plot_data(x: List[float], y: List[float], color: str = 'blue') -> mpl_figure.Figure:
    """Plots x and y data on a new figure."""
    fig, ax = plt.subplots()
    ax.plot(x, y, color=color)
    return fig

5. Virtual Environments & Packaging

Always use virtual environments (venv, conda) for dependency management. Pin Matplotlib to a specific stable version in requirements.txt or pyproject.toml to ensure reproducibility.

✅ GOOD (requirements.txt):

matplotlib==3.10.0
numpy==1.26.0
pandas==2.1.0
seaborn==0.13.0

6. Testing Approaches

Integrate ruff check --fix into your CI/CD pipeline. Add specific tests for new features and bug fixes, especially image comparison tests for visual changes.

✅ GOOD (Example test structure):

# tests/test_plotting.py
import matplotlib.pyplot as plt
import numpy as np
from my_module import generate_plot # Assuming generate_plot is in my_module

def test_generate_plot_output():
    data = np.array([1, 2, 3, 4, 5])
    fig = generate_plot(data)
    assert isinstance(fig, plt.Figure)
    assert len(fig.axes) == 1
    # Add more assertions, e.g., check title, labels, data points
    plt.close(fig) # Clean up figure after test

# For visual regression testing (requires specific tooling like pytest-mpl)
# def test_plot_visual_regression(mpl_image_compare):
#     fig = generate_plot(np.array([1, 5, 2, 8]))
#     assert mpl_image_compare(fig, "expected_plot_output")