Zappy - distributed processing with NumPy and Zarr

Zappy is for distributed processing of chunked NumPy arrays on engines like Pywren, Apache Spark, and Apache Beam.

The zappy.base module defines a ZappyArray class that exposes the same interface as numpy.ndarray, and which is backed by distributed storage and processing. The array is broken into chunks, and is typically loaded from Zarr, and each chunk is processed independently.

There are a few engines provided:

• direct - for eager in-memory processing
• spark - for processing using Spark
• beam - for processing using Beam or Google Dataflow
• executor - for processing using Python's concurrent.futures.Executor, of which Pywren is a notable implementation

Beam currently only runs on Python 2.

Full coverage of the numpy.ndarray interface is not provided. Only enough has been implemented to support running parts of Scanpy, as demonstrated in the Single Cell Experiments repo.

Installation

pip install zappy

Alternatively, zappy can be installed using Conda (most easily obtained via the Miniconda Python distribution):

conda install -c conda-forge zappy

Demo

Take a look at the rendered demo Jupyter notebook, or try it out yourself as follows.

Create and activate a Python 3 virtualenv, and install the requirements:

python3 -m venv venv
. venv/bin/activate
pip install -r requirements.txt
pip install -e .
pip install s3fs jupyter

Then run the notebook with:

jupyter notebook demo.ipynb

Testing

There is a test suite for all the engines, covering both Python 2 and 3.

Run everything in one go with tox:

pip install tox
tox

Formatting:

pip install black
black zappy tests/* *.py

Coverage:

pip install pytest-cov
pytest --cov-report html --cov=zappy
open htmlcov/index.html

Publishing

pip install twine
python setup.py sdist
twine upload -r pypi dist/zappy-0.1.0.tar.gz

If successful, the package will be available on PyPI.