pysz

A user-friendly and easy-to-maintain python wrapper for SZ3 using a python-to-c package Cython>=3.0.10.
Pysz is fully based on SZ3(https://github.com/szcompressor/SZ3). The pure SZ3 C++ library is incorporated into pysz interface through Cython, so the compression && decompression function is running much faster than Python code.

How to use pysz

We provided a test code in ./test/test.py as shown below:

from pysz import sz

# setup configuration of compression
dim = (101, 203869, 3)
compressor = sz.sz(*dim)
filename = 'traj_xyz.dat'
datatype = 'float'

# compression
cmpData3, cmp_ratio_2 = compressor.compress(datatype, 'sz3.config', filename)

# decompression
decmpData2 = compressor.decompress(datatype, 'sz3.config', filename + '.sz')

# verification
compressor.verify(datatype, 'sz3.config', filename, filename + '.sz')

The input data can either be the file path to the data file or numpy array contain the data.

Requirements

Python>=3.10

Installation

Method 1. install from pypi

pip install pysz

Without the support of your current environment, it could take a few more time at the step of "Building wheels for collected packages...".
If you are running a linux system and using x86 or x64 CPU platform, pysz can be installed directly from a pre-compiled .whl file

Method 2. download source code and compile locally

First, download the source code from github rep.

git clone https://github.com/JasonYang60/pysz.git

Second, make sure cmake && gcc is installed.
Pysz uses python-to-c package Cython>=3.0.10, so what setup.py does includes downloading c++ source code from github and automatically applying cmake && make command to finish c++ code compilation.
Then

pip install .

Generate pre-compiled .whl and .tar.gz file

To pack this project up with .whl file, some python setup tool package (wheel, build) and are required.

pip install wheel
pip install build

Run following code to generate .whl and .tar.gz in ./dist:

python -m build

tip: there is still some issues remaining to be fixed. This command will also work:

python setup.py sdist bdist_wheel

Launch in docker

Dockerfile is provided for testing.

docker build --name pysz .
docker run -it --name pysz_container pysz /bin/bash

Then in container shell:

pip install .
cd test
python ./test.py

TODO list:

to test openMP and other utils && features
more data types to be supported