pls — Pretty Little Summary

A DataFrame, at a glance

Nulls, dtypes, and per-column stats—no .describe() ceremony.

Input

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

Output (result.content)

A pandas DataFrame with 3 rows and 3 columns. Nulls: 0. Memory: 225.0 B. Columns: product (str, cardinality: 3 categories including 'Widget', 'Gadget', 'Doohickey'), price (float64, stats: range 20 to 40, mean 30, std 10), quantity (int64, stats: range 50 to 1e+02, mean 75, std 25). Sample row: {'product': 'Widget', 'price': '19.99', 'quantity': '100'}.

A messy folder, sorted out

Point it at a directory and get every file described—no unzip-and-poke required.

Input

pls.describe("experiment_042/")

Output (result.content)

experiment_042/ (4 files, 1 subdirectories)
├── figures/
│   ├── confusion_matrix.png - A PIL image 1024x768 in RGB mode.
│   └── loss_curve.png - A PIL image 1024x768 in RGB mode.
└── results.csv - A pandas DataFrame with 5 rows and 2 columns. Nulls: 0. Memory: 212.0 B. Columns: trial (int64, stats: range 1 to 5, mean 3, std 1.6), score (float64, stats: range 0.77 to 0.9, mean 0.82, std 0.052). Sample row: {'trial': '1', 'score': '0.81'}.

A file you haven't opened yet

Pass a path—it reads the parquet file and profiles it like a live DataFrame.

Input

pls.describe("sensor_readings.parquet")

Output (result.content)

A pandas DataFrame with 5 rows and 3 columns. Nulls: 1. Memory: 265.0 B. Columns: reading (float64, 1 nulls, stats: range 20 to 23, mean 22, std 1.4), sensor_id (int64, stats: range 1 to 5, mean 3, std 1.6), status (str, cardinality: 2 categories including 'ok', 'error'). Sample row: {'sensor_id': '1', 'reading': '21.4', 'status': 'ok'}.

An unfitted model, explained

Not <Pipeline object at 0x...>—the actual steps and what it expects.

Input

pipe = Pipeline([
    ("imputer", SimpleImputer()),
    ("scaler", StandardScaler()),
])
pls.describe(pipe)

Output (result.content)

A unfitted sklearn Pipeline with 2 steps:
1. 'imputer': SimpleImputer
2. 'scaler': StandardScaler
Expects input shape (*, ?), outputs class predictions.

More adapters, more inputs—browse the full playground.

Quick start

Install

Zero required dependencies. Opt into extras when you're ready.

pip install pretty-little-summary

Usage

One function. Pass it anything. It figures out the rest.

Corrected manners.py

import pretty_little_summary as pls

messy_data = load_nightmare_csv()

print(messy_data) # prints too muchrude.

messy_data.info() # buries the pointtoo much drama.

result = pls.describe(messy_data)much better.

print(result.content)

# A pandas DataFrame with 12,481 rows and 9 columns. Nulls: 342.

# Columns: order_id, customer, region, total, ...

What you get back

A Summary with human-readable content and structured metadata.

result.content

pandas.DataFrame | Shape: (3, 3)
Columns: product, price, quantity

Structured metadata

result.meta
# {"object_type": "pandas.DataFrame",
#  "shape": [3, 3],
#  "columns": ["product", "price", "quantity"]}

Summary object

.content str — formatted summary

.meta dict — structured metadata

.history list — jupyter code lineage

Jupyter history

In notebooks, PLS traces the code that created your object. A tiny time machine.

result = pls.describe(df_filtered)
print(result.history)

# ['df = pd.read_csv("data.csv")',
#  'df_clean = df.dropna()',
#  'df_filtered = df_clean[df_clean["value"] > 100]']