sinaplot

sinaplot draws SinaPlots from pandas data with Matplotlib. A SinaPlot shows every observation while using the width of the point distribution to represent the estimated density.

This implementation has two features intended to make comparisons more truthful:

• The area of each violin is proportional to the number of observations in that group. A larger group therefore has a larger envelope instead of every violin being normalized independently.
• jitter_method="even" spreads nearby points in display space, reducing overlap and producing a more even visual density.

Installation

pip install sinaplot

Python 3.10 or later is required.

Quick start

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

from sinaplot import sinaplot

rng = np.random.default_rng(7)
data = pd.DataFrame(
    {
        "condition": np.repeat(["Control", "Treatment A", "Treatment B"], [50, 90, 130]),
        "response": np.concatenate(
            [
                rng.normal(0.0, 0.8, 50),
                rng.normal(0.7, 1.0, 90),
                np.r_[rng.normal(-0.2, 0.45, 65), rng.normal(1.7, 0.55, 65)],
            ]
        ),
    }
)

fig, ax = plt.subplots(figsize=(7, 4.5))
sinaplot(
    x="condition",
    y="response",
    data=data,
    jitter_method="even",
    random_seed=42,
    ax=ax,
)
fig.tight_layout()
plt.show()

sinaplot returns the Matplotlib Axes, so normal Matplotlib customization continues to work:

ax = sinaplot(
    "condition",
    "response",
    data,
    order=["Treatment B", "Treatment A", "Control"],
    color="tab:purple",
    violin_kws={"alpha": 0.2},
    scatter_kws={"edgecolor": "white", "linewidth": 0.3},
)
ax.set_title("Response by condition")

Main options

Rows with missing categories or missing/non-finite numerical values are omitted. Groups with fewer than two distinct values cannot have a kernel density estimate and are represented by a horizontal line.

How area scaling works

For each group, the displayed kernel density is numerically normalized over the plotted range and then multiplied by the number of finite observations in that group. All groups use one common horizontal scale. Consequently, violin area is proportional to sample size, while the local width still represents density.

The points are always constrained to the corresponding density envelope. With jitter_method="even", an iterative repulsion calculation in display coordinates gives nearby points more uniform spacing. Set random_seed when you need exactly reproducible coordinates.

Development

git clone https://github.com/okumuralab/sinaplot.git
cd sinaplot
python -m pip install -e ".[test]"
python -m pytest

Build the wheel and source distribution with:

python -m pip install build
python -m build

Reference

Sidiropoulos N, Sohi SH, Pedersen TL, Porse BT, Winther O, Rapin N, Bagger FO. “SinaPlot: an enhanced chart for simple and truthful representation of single observations over multiple classes.” Journal of Computational and Graphical Statistics 27(3), 673–676 (2018). doi:10.1080/10618600.2017.1366914. Preprint: doi:10.1101/028191.

This project is an independent Python implementation and is not affiliated with the authors of the paper or the R packages that implement SinaPlot.

License

MIT