Tuplex: Blazing Fast Python Data Science

Website Documentation

Tuplex is a parallel big data processing framework that runs data science pipelines written in Python at the speed of compiled code. Tuplex has similar Python APIs to Apache Spark or Dask, but rather than invoking the Python interpreter, Tuplex generates optimized LLVM bytecode for the given pipeline and input data set. Under the hood, Tuplex is based on data-driven compilation and dual-mode processing, two key techniques that make it possible for Tuplex to provide speed comparable to a pipeline written in hand-optimized C++.

You can join the discussion on Tuplex on our Gitter community or read up more on the background of Tuplex in our SIGMOD'21 paper.

Contributions welcome!

Contents

• Installation
  • Docker image
  • Pypi
• Building
  • MacOS build from source
  • Ubuntu build from source
  • Customizing the build
• Example
• License

Installation

To install Tuplex, you can use a PyPi package for Linux, or a Docker container for MacOS which will launch a jupyter notebook with Tuplex preinstalled.

Docker

docker run -p 8888:8888 tuplex/tuplex

PyPI

pip install tuplex

Building

Tuplex is available for MacOS and Linux. The current version has been tested under MacOS 10.13-10.15 and Ubuntu 18.04 and 20.04 LTS. To install Tuplex, simply install the dependencies first and then build the package.

MacOS build from source

To build Tuplex, you need several other packages first which can be easily installed via brew.

brew install [email protected] boost boost-python3 aws-sdk-cpp pcre2 antlr4-cpp-runtime googletest gflags yaml-cpp celero
python3 -m pip cloudpickle numpy
python3 setup.py install

Ubuntu build from source

To faciliate installing the dependencies for Ubuntu, we do provide two scripts (scripts/ubuntu1804/install_reqs.sh for Ubuntu 18.04, or scripts/ubuntu2004/install_reqs.sh for Ubuntu 20.04). To create an up to date version of Tuplex, simply run

./scripts/ubuntu1804/install_reqs.sh
python3 -m pip cloudpickle numpy
python3 setup.py install

Customizing the build

Besides building a pip package, cmake can be also directly invoked. To compile the package via cmake

mkdir build
cd build
cmake ..
make -j$(nproc)

The python package corresponding to Tuplex can be then found in build/dist/python with C++ test executables based on googletest in build/dist/bin.

To customize the cmake build, the following options are available to be passed via -D=:

For example, to create a debug build which outputs PDFs use the following snippet:

cmake -DCMAKE_BUILD_TYPE=Debug -DGENERATE_PDFS=ON ..

Example

Tuplex can be used in python interactive mode, a jupyter notebook or by copying the below code to a file. To try it out, run the following example:

from tuplex import *
c = Context()
res = c.parallelize([1, 2, None, 4]).map(lambda x: (x, x * x)).collect()
# this prints [(1, 1), (2, 4), (4, 16)]
print(res)

More examples can be found here.

License

Tuplex is available under Apache 2.0 License, to cite the paper use:

@inproceedings{10.1145/3448016.3457244,
author = {Spiegelberg, Leonhard and Yesantharao, Rahul and Schwarzkopf, Malte and Kraska, Tim},
title = {Tuplex: Data Science in Python at Native Code Speed},
year = {2021},
isbn = {9781450383431},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3448016.3457244},
doi = {10.1145/3448016.3457244},
booktitle = {Proceedings of the 2021 International Conference on Management of Data},
pages = {1718–1731},
numpages = {14},
location = {Virtual Event, China},
series = {SIGMOD/PODS '21}
}

(c) 2017-2021 Tuplex contributors