Pretty Little Summary

Automatic structured summaries of Python objects — DataFrames, arrays, models, plots, and more.

Install

pip install pretty-little-summary

Optional adapters are enabled automatically when their libraries are installed.

No install? Try it in the browser: live playground (runs via Pyodide, nothing uploaded).

Features

• Single function API: pls.describe(obj)
• 40+ adapters across data, viz, and ML libraries
• Works with built-ins out of the box (no required deps)
• Jupyter/IPython history capture for better context
• Deterministic, bounded question-focused tabular views with pluggable relevance scorers

Quick Start

import pretty_little_summary as pls
import pandas as pd

df = pd.DataFrame({
    "product": ["Widget", "Gadget", "Doohickey"],
    "price": [19.99, 29.99, 39.99],
    "quantity": [100, 50, 75]
})

result = pls.describe(df)
print(result.content)
print(result.meta)

# Rank all columns for a question without changing the full profile.
focused = pls.focus_profile(
    result,
    "How does price relate to quantity?",
    max_columns=8,
    max_chars=4000,
)
print(focused.content)

Built-in Types

import pretty_little_summary as pls

print(pls.describe([1, 2, 3]).content)
print(pls.describe({"name": "Alice", "age": 30}).content)

NumPy Arrays

import numpy as np
import pretty_little_summary as pls

arr = np.random.rand(100, 50)
result = pls.describe(arr)
print(result.content)

Pandas DataFrames

import pandas as pd
import pretty_little_summary as pls

df = pd.read_csv("data.csv")
result = pls.describe(df)
print(result.content)

Files and dynamic imports

File paths are detected through a shared capability registry using extensions and, where available, magic bytes. In native Python, dynamic_imports=True imports matching optional libraries that are already installed; it never runs a package manager or accesses the network:

import pretty_little_summary as pls

result = pls.describe("measurements.h5", dynamic_imports=True)
print(result.content)

# Hosts that manage packages themselves can inspect the same requirements:
print(pls.requirements_for_path("measurements.h5"))

The browser playground uses these requirements to install missing Pyodide or micropip packages automatically, so there is no package-tier selector.

Matplotlib Figures

import matplotlib.pyplot as plt
import pretty_little_summary as pls

fig, ax = plt.subplots()
ax.plot([1, 2, 3], [4, 5, 6])
result = pls.describe(fig)
print(result.content)

History Tracking (Jupyter/IPython)

When running inside Jupyter, pretty_little_summary can capture recent code history that created your object:

import pandas as pd
import pretty_little_summary as pls

df = pd.read_csv("data.csv")
df_clean = df.dropna()
result = pls.describe(df_clean)
print(result.history)

Docs site

The docs (docs/index.html, docs/playground.html, docs/eval-report.html) are a static site with no build step. To run it locally:

cd docs && python3 -m http.server 8000

Then open http://localhost:8000. Notes:

• playground.html loads Pyodide from a CDN, so it needs internet access even when served locally.
• playground.html installs pretty-little-summary from docs/dist/*.whl so it always matches this commit's src/ instead of a possibly-stale PyPI release. That wheel isn't committed — CI builds it on every docs deploy via python -m build --wheel -o docs/dist. Build it yourself before serving locally, or the playground falls back to installing from PyPI:

  .venv/bin/python -m build --wheel -o docs/dist
  

• eval-report.html isn't committed — generate it first with:

  .venv/bin/python -m evals.runner fetch
  .venv/bin/python -m evals.runner run
  .venv/bin/python -m evals.viewer --out docs/eval-report.html
  

  (CI does this automatically on every docs deploy.) Its markup lives in the committed docs/eval-report-template.html; evals/viewer.py only injects run data into it.
• The playground's curated/gallery example code comes from docs/examples/*.txt, regenerated via:

  .venv/bin/python scripts/export_examples.py
  

Troubleshooting

ModuleNotFoundError: No module named 'pretty_little_summary'

• Ensure you installed the package in the current environment.
• Restart your kernel or interpreter.

Missing optional libraries

If an adapter isn’t available, install its library:

pip install pandas numpy matplotlib

Or install all optional dependencies:

pip install pretty-little-summary[all]