Skip to content

Latest commit

 

History

History
72 lines (65 loc) · 4.02 KB

paper.md

File metadata and controls

72 lines (65 loc) · 4.02 KB
Raw
title tags authors affiliations date bibliography
Hickle: A HDF5-based python pickle replacement
Python
astronomy
name orcid affiliation
Danny C. Price
0000-0003-2783-1608
1, 2
name orcid affiliation
Ellert van der Velden
0000-0002-1559-9832
2
name orcid affiliation
Sébastien Celles
0000-0001-9987-4338
3
name orcid affiliation
Pieter T. Eendebak
0000-0001-7018-1124
4, 5
name orcid affiliation
Michael M. McKerns
0000-0001-8342-3778
6
name affiliation
Eben M. Olson
7
name affiliation
Colin Raffel
8
name affiliation
Bairen Yi
9
name affiliation
Elliott Ash
10
name index
Department of Astronomy, University of California Berkeley, Berkeley CA 94720
1
name index
Centre for Astrophysics & Supercomputing, Swinburne University of Technology, Hawthorn, VIC 3122, Australia
2
name index
Thermal Science and Energy Department, Institut Universitaire de Technologie de Poitiers - Université de Poitiers, France
3
name index
QuTech, Delft University of Technology, P.O. Box 5046, 2600 GA Delft, The Netherlands
4
name index
Netherlands Organisation for Applied Scientific Research (TNO), P.O. Box 155, 2600 AD Delft, The Netherlands
5
name index
Institute for Advanced Computational Science, Stony Brook University, Stony Brook, NY 11794-5250
6
name index
Department of Laboratory Medicine, Yale University, New Haven CT 06510 USA
7
name index
Google Brain, Mountain View, CA, 94043
8
name index
The Hong Kong University of Science and Technology
9
name index
ETH Zurich
10
10 November 2018
paper.bib

Summary

hickle is a Python 2/3 package for quickly dumping and loading python data structures to Hierarchical Data Format 5 (HDF5) files [@hdf5]. When dumping to HDF5, hickle automatically convert Python data structures (e.g. lists, dictionaries, numpy arrays [@numpy]) into HDF5 groups and datasets. When loading from file, hickle automatically converts data back into its original data type. A key motivation for hickle is to provide high-performance loading and storage of scientific data in the widely-supported HDF5 format.

hickle is designed as a drop-in replacement for the Python pickle package, which converts Python object hierarchies to and from Python-specific byte streams (processes known as 'pickling' and 'unpickling' respectively). Several different protocols exist, and files are not designed to be compatible between Python versions, nor interpretable in other languages. In contrast, hickle stores and loads files from HDF5, for which application programming interfaces (APIs) exist in most major languages, including C, Java, R, and MATLAB.

Python data structures are mapped into the HDF5 abstract data model in a logical fashion, using the h5py package [@collette:2014]. Metadata required to reconstruct the hierarchy of objects, and to allow conversion into Python objects, is stored in HDF5 attributes. Most commonly used Python iterables (dict, tuple, list, set), and data types (int, float, str) are supported, as are numpy N-dimensional arrays. Commonly-used astropy data structures and scipy sparse matrices are also supported.

hickle has been used in many scientific research projects, including:

  • Visualization and machine learning on volumetric fluorescence microscopy datasets from histological tissue imaging [@Durant:2017].
  • Caching pre-computed features for MIDI and audio files for downstream machine learning tasks [@Raffel:2016].
  • Storage and transmission of high volume of shot-gun proteomics data, such as mass spectra of proteins and peptide segments [@Zhang:2016].
  • Storage of astronomical data and calibration data from radio telescopes [@Price:2018].

hickle is released under the MIT license, and is available from PyPi via pip; source code is available at https://github.com/telegraphic/hickle.

References