tab_err

Inject Realistic Errors Into Tables

📊 🔎 ✅

Test how your data pipelines and ML models react to tabular data that contains realistic errors.

tab_err injects realistic errors into tabular data such as database tables and DataFrames. The library is developed and maintained by the Cognitive Algorithms Lab at BHT Berlin.

Using error-free tables as input, tab_err lets users define an error model that perturbs the table and can be shared as metadata. Researchers and data practitioners can generate errors in a controlled way, evaluate how their systems behave, and exchange error scenarios reproducibly.

How it Works

The library's building blocks are ErrorMechanisms, ErrorTypes, and ErrorModels.

• An ErrorMechanism describes the error's distribution - that's where incorrect cells appear in the table. We support erroneous at random (EAR), erroneous not at random (ENAR) and erroneous completely at random (ECAR).
• An ErrorType describes how the value is wrong: a typo, an outlier, a category swap, and so on. Read the documentation for a full list of supported error types.
• An ErrorModel is a set of mechanisms and types to perturb existing data with realistic errors. It is shareable as metadata.

tab_err is supported by a pandas backend.

Examples

from sklearn.datasets import load_iris

from tab_err import error_type
from tab_err.api import high_level

df = load_iris(as_frame=True).frame
corrupted_df, error_mask = high_level.create_errors(
    data=df,
    error_rate=0.5,
    error_types_to_exclude=[error_type.MissingValue()],
    seed=42,
)
print("Original:")
print(df.head(2).to_string(index=False))

print("\nCorrupted:")
print(corrupted_df.head(2).to_string(index=False))

print("\nCorrupted cells:", int(error_mask.to_numpy().sum()))

Example output:

Original:
 sepal length (cm)  sepal width (cm)  petal length (cm)  petal width (cm)  target
               5.1               3.5                1.4               0.2       0
               4.9               3.0                1.4               0.2       0

Corrupted:
 sepal length (cm)  sepal width (cm)  petal length (cm)  petal width (cm)   target
               5.1              35.0           1.400000         -2.775759 0.420326
               4.9              30.0           1.820326         -3.087558 0.000000

Corrupted cells: 375

For a detailed guide and more examples, see our Getting Started Notebook and the documentation.

Where to get it

The source code is hosted on GitHub at https://github.com/calgo-lab/tab_err. Binary installers for the latest releases are available at the Python Package Index (PyPI) https://pypi.org/project/tab-err.

# with pip
pip install tab-err

# with uv
uv add tab-err

Contributing

To develop tab_err, install the uv package manager. Run tests with uv run pytest. Develop on feature branches and open pull requests when you're ready. Make sure that your changes are tested, documented, and clearly described in the pull request.

Citation

If you use the error model that's underlying tab_err for a scientific publication, we would appreciate your citation.

@article{10.1145/3774914,
author = {Jung, Philipp and J\"{a}ger, Sebastian and Chandler, Nicholas and Biessmann, Felix},
title = {Towards Realistic Error Models for Tabular Data},
year = {2025},
issue_date = {December 2025},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {17},
number = {4},
issn = {1936-1955},
url = {https://doi.org/10.1145/3774914},
doi = {10.1145/3774914},
journal = {J. Data and Information Quality},
month = dec,
articleno = {28},
numpages = {27},
keywords = {Tabular data, data quality, data errors, data error generation, error model, realistic error model, error type}
}