Hickle is an HDF5 based clone of
pickle, with a twist: instead of serializing to a pickle file,
Hickle dumps to an HDF5 file (Hierarchical Data Format). It is designed to be a "drop-in" replacement for pickle (for common data objects), but is
really an amalgam of
pickle with extended functionality.
hickle is a neat little way of dumping python variables to HDF5 files that can be read in most programming
languages, not just Python. Hickle is fast, and allows for transparent compression of your data (LZF / GZIP).
hickle is designed to be a drop-in replacement for
pickle (or something like
json), it works very differently.
Instead of serializing / json-izing, it instead stores the data using the excellent h5py module.
The main reasons to use hickle are:
The main reasons not to use hickle are:
So, if you want your data in HDF5, or if your pickling is taking too long, give hickle a try.
Hickle is particularly good at storing large numpy arrays, thanks to
h5py running under the hood.
Documentation for hickle can be found at telegraphic.github.io/hickle/.
Hickle is nice and easy to use, and should look very familiar to those of you who have pickled before.
import os import hickle as hkl import numpy as np # Create a numpy array of data array_obj = np.ones(32768, dtype='float32') # Dump to file hkl.dump(array_obj, 'test.hkl', mode='w') # Dump data, with compression hkl.dump(array_obj, 'test_gzip.hkl', mode='w', compression='gzip') # Compare filesizes print('uncompressed: %i bytes' % os.path.getsize('test.hkl')) print('compressed: %i bytes' % os.path.getsize('test_gzip.hkl')) # Load data array_hkl = hkl.load('test_gzip.hkl') # Check the two are the same file assert array_hkl.dtype == array_obj.dtype assert np.all((array_hkl, array_obj))
A major benefit of
pickle is that it allows fancy HDF5 features to
be applied, by passing on keyword arguments on to
h5py. So, you can do things like:
hkl.dump(array_obj, 'test_lzf.hkl', mode='w', compression='lzf', scaleoffset=0, chunks=(100, 100), shuffle=True, fletcher32=True)
A detailed explanation of these keywords is given at http://docs.h5py.org/en/latest/high/dataset.html, but we give a quick rundown below.
In HDF5, datasets are stored as B-trees, a tree data structure that has speed benefits over contiguous
blocks of data. In the B-tree, data are split into chunks,
which is leveraged to allow dataset resizing and
compression via filter pipelines. Filters such as
scaleoffset move your data around to improve compression ratios, and
fletcher32 computes a checksum.
These file-level options are abstracted away from the data model.
Hickle runs a lot faster than pickle with its default settings, and a little faster than pickle with
In : import numpy as np In : x = np.random.random((2000, 2000)) In : import pickle In : f = open('foo.pkl', 'w') In : %time pickle.dump(x, f) # slow by default CPU times: user 2 s, sys: 274 ms, total: 2.27 s Wall time: 2.74 s In : f = open('foo.pkl', 'w') In : %time pickle.dump(x, f, protocol=2) # actually very fast CPU times: user 18.8 ms, sys: 36 ms, total: 54.8 ms Wall time: 55.6 ms In : import hickle In : f = open('foo.hkl', 'w') In : %time hickle.dump(x, f) # a bit faster dumping <type 'numpy.ndarray'> to file <HDF5 file "foo.hkl" (mode r+)> CPU times: user 764 us, sys: 35.6 ms, total: 36.4 ms Wall time: 36.2 ms
So if you do continue to use pickle, add the
protocol=2 keyword (thanks @mrocklin for pointing this out).
For storing python dictionaries of lists, hickle beats the python json encoder, but is slower than uJson. For a dictionary with 64 entries, each containing a 4096 length list of random numbers, the times are:
json took 2633.263 ms uJson took 138.482 ms hickle took 232.181 ms
It should be noted that these comparisons are of course not fair: storing in HDF5 will not help you convert something into JSON, nor will it help you serialize a string. But for quick storage of the contents of a python variable, it's a pretty good option.
pip by running
pip install hickle from the command line.
You should have Python 3.5 and above installed
Install h5py (Official page: http://docs.h5py.org/en/latest/build.html)
Install hdf5 (Official page: http://www.hdfgroup.org/ftp/HDF5/current/src/unpacked/release_docs/INSTALL)
via terminal: git clone https://github.com/telegraphic/hickle.git
via manual download: Go to https://github.com/telegraphic/hickle and on right hand side you will find
Download ZIP file
cd to your downloaded
Then run the following command in the
python setup.py install
Once installed from source, run
python setup.py test to check it's all working.
Contributions and bugfixes are very welcome. Please check out our contribution guidelines for more details on how to contribute to development.
Price et al., (2018). Hickle: A HDF5-based python pickle replacement. Journal of Open Source Software, 3(32), 1115, https://doi.org/10.21105/joss.01115