Python and Data Science .gitattributes

Python / Data Science .gitattributes

Python data science projects combine source code, Jupyter notebooks, large datasets, and trained model files. Each category needs different Git handling.

Recommended Configuration

# Auto detect
* text=auto

# Python source
*.py     text diff=python
*.pyx    text diff=python
*.pxd    text diff=python
*.pyi    text diff=python

# Config / packaging
*.cfg    text
*.ini    text
*.toml   text
*.yaml   text
*.yml    text
setup.py text diff=python
pyproject.toml text

# Jupyter notebooks
*.ipynb  text -diff

# Lock files
poetry.lock   text -diff
Pipfile.lock  text -diff
requirements.txt text

# Data files (text-based)
*.csv    text
*.tsv    text
*.json   text
*.jsonl  text
*.xml    text

# Data files (binary)
*.pkl    binary
*.pickle binary
*.parquet binary
*.feather binary
*.hdf5   binary
*.h5     binary
*.npy    binary
*.npz    binary
*.arrow  binary

# Trained models
*.pt     binary
*.pth    binary
*.onnx   binary
*.pb     binary
*.tflite binary
*.joblib binary

# Python compiled
*.pyc    binary
*.pyd    binary
*.so     binary
*.egg    binary
*.whl    binary

# Images / plots
*.png    binary
*.jpg    binary
*.jpeg   binary
*.svg    text

# Shell scripts
*.sh     text eol=lf

Jupyter Notebooks and -diff

Jupyter .ipynb files are JSON with embedded outputs (images encoded as base64, HTML tables, etc.). Standard diffs of notebooks are nearly unreadable. Using text -diff ensures line ending normalization while suppressing noisy diffs.

For better notebook diffs, consider tools like nbdime:

# In .gitconfig
[diff "jupyternotebook"]
  command = git-nbdiffdriver diff
[merge "jupyternotebook"]
  command = git-nbmergedriver merge %O %A %B %L %P

# In .gitattributes (with nbdime)
*.ipynb text diff=jupyternotebook merge=jupyternotebook

Model and Data Files

Trained models (.pt, .onnx, .pb) and data files (.parquet, .hdf5) are binary. For large models, consider Git LFS:

*.pt  filter=lfs diff=lfs merge=lfs -text
*.h5  filter=lfs diff=lfs merge=lfs -text