squirrel-datasets-core is an extension of the Squirrel library. squirrel-datasets-core is a hub where the user can

1. explore existing public datasets registered in the data mesh and load them with the ease and speed of squirrel
2. preprocess their datasets and share them with other users.

For preprocessing, we currently support Spark as the main tool to carry out the task.

If you have any questions or would like to contribute, join our Slack community!

Install squirrel-core and squirrel-datasets-core with pip. Note that you can install with different dependencies based on your requirements for squirrel drivers. For using the Torchvision driver call:

pip install "squirrel-core[torch]"
pip install "squirrel-datasets-core[torchvision]"

For using the Huggingface or Deeplake driver call:

pip install "squirrel-datasets-core[huggingface]"
pip install "squirrel-datasets-core[deeplake]"

For using the Spark preprocessing pipelines call:

pip install "squirrel-datasets-core[preprocessing]"

If you would like to get Squirrel's full functionality, install squirrel-core and squirrel-datasets-core with all their dependencies.

pip install "squirrel-core[all]"
pip install "squirrel-datasets-core[all]"

A great feature of squirrel-datasets-core is that you can easily load data from common databases such as Huggingface, Activeloop Deeplake, Hub and Torchvision with one line of code. And you get to enjoy all of Squirrel’s benefits for free! Check out the documentation on how to interface with these libraries.

from squirrel_datasets_core.driver.huggingface import HuggingfaceDriver

it = HuggingfaceDriver("cifar100").get_iter("train").filter(custom_filter).map(custom_augmentation)

# your train loop
for item in it:
  out = model(item)
  # ...

Visit our documentation on Readthedocs.

squirrel-datasets-core is open source and community contributions are welcome!

Check out the contribution guide to learn how to get involved. Please follow our recommendations for best practices and code style.

We are Merantix Momentum, a team of ~30 machine learning engineers, developing machine learning solutions for industry and research. Each project comes with its own challenges, data types and learnings, but one issue we always faced was scalable data loading, transforming and sharing. We were looking for a solution that would allow us to load the data in a fast and cost-efficient way, while keeping the flexibility to work with any possible dataset and integrate with any API. That's why we build Squirrel – and we hope you'll find it as useful as we do! By the way, we are hiring!

If you use Squirrel Datasets in your research, please cite Squirrel using:

@article{2022squirrelcore,
  title={Squirrel: A Python library that enables ML teams to share, load, and transform data in a collaborative, flexible, and efficient way.},
  author={Squirrel Developer Team},
  journal={GitHub. Note: https://github.com/merantix-momentum/squirrel-core},
  year={2022}
}