datascore

ML readiness scoring for tabular datasets.

Point it at a DataFrame and get a structured report telling you whether your data is ready for ML training — and if not, exactly why and in what order to fix it.

Install

pip install datascore

Usage

import seaborn as sns
from datascore import score

df = sns.load_dataset("titanic")
report = score(df, target="survived")
report.show()

datascore infers classification versus regression from the target. For ambiguous integer targets, pass task="classification" or task="regression" explicitly.

Output

datascore Report

Rows: 891 | Features: 15 | Target: survived
Score: 45/100 — NOT READY

BLOCKERS

- age: 19.9% missing values
- deck: 77.2% missing values

WARNINGS

- Missing values detected: 6.5% overall
- 107 duplicate rows detected
- High skew in sibsp: 3.6891
- High skew in parch: 2.7445
- High skew in fare: 4.7793

INFO

- Outliers in age: 11 rows
- Outliers in sibsp: 46 rows
- Outliers in parch: 213 rows
- Outliers in fare: 116 rows
- No constant features detected
- No infinite values detected
- Class balance: 62/38

Save report to markdown

report.save("report.md")

What it checks

Scoring

Starts at 100. Each blocker deducts 15 points, each warning deducts 5.

Why not Great Expectations or Pandera?

Those tools validate data against rules you define upfront.

datascore requires no configuration — it tells you what the problems are without you having to know what to look for first.

Assessment, not validation.

License

MIT