shirabe-profiling

A fast, Polars-native data profiling tool that generates interactive HTML reports equipped with automated data quality alerts to streamline your exploratory data analysis.

Quickstart | Report tour | Use it in code | License

shirabe-profiling reads a Polars DataFrame and writes one HTML report that you open in a browser. The report shows the facts of each column, flags quality problems as alerts, and explains how to fix each alert.

• Fast, the engine is Polars, and overview mode skips the slow parts for a quick first look. See the benchmarks.
• A guide for every alert, the Guide tab covers all 12 alerts, with each limit, and with some possible fixes and the sources for most of them.
• Made for modeling, the target section shows how columns link to the target and warns about leakage, and skewed columns get log1p and Yeo-Johnson tabs.
• Easy to use in code, read the numbers with .to_dict(), .to_json(), .stats, and .alerts, and the package is fully typed.
• Easy on the eyes, the report follows the theme of your system, and you can switch between dark and light in the settings menu.
• Open to many inputs, Polars frames, pandas frames, CSV paths, and Parquet paths all work.

Quickstart

Install the package from PyPI:

pip install shirabe-profiling

With uv, run uv add shirabe-profiling in a project, or uv pip install shirabe-profiling in a virtual environment.

For pandas tables with text columns, install the extra: pip install "shirabe-profiling[pandas]".

Then write a report:

import polars as pl
from shirabe_profiling import ProfileReport

df = pl.read_csv("data.csv")
report = ProfileReport(df)
report.to_html("report.html")

You can name the column that you want to predict to add a target section:

report = ProfileReport(df, target="price")

The report loads ECharts and Tailwind CSS from a CDN, so it needs an internet connection to display correctly.

Report tour

Click any picture to open it full size.

Target

Column details

Correlations

Guide

The report also has:

• A correlation heatmap for each method (Phik, Pearson, Spearman, Cramér's V, and Kendall), with a list of the strongest pairs.
• An Interactions section with scatter plots, box plots, and a column picker.
• A Missing Patterns table and an Outliers table.
• A Sample section with the first rows, the last rows, random rows, and the rows with the most gaps.

Use it in code

report = ProfileReport(df, title="Home Prices Profile")

stats = report.to_dict()
alerts = report.alerts
column_stats = report.stats["area_m2"]
correlations = report.correlations

from shirabe_profiling import ProfileReport, ProfileConfig

config = ProfileConfig(
    title="Home Prices Profile",
    mode="overview",
    missing_threshold=0.1,
    histogram_bins=30,
)
report = ProfileReport(df, config=config)

License

This project is a derivative work based on dataxid-profiling by DataXID, originally licensed under the Apache License 2.0.

• LICENSE, the Apache-2.0 grant for the whole repository
• LICENSES/, the full text of each license that the files use
• REUSE.toml, the license and the copyright holder of each file