bcolz: columnar and compressed data containers

bcolz provides columnar, chunked data containers that can be compressed either in-memory and on-disk. Column storage allows for efficiently querying tables, as well as for cheap column addition and removal. It is based on NumPy, and uses it as the standard data container to communicate with bcolz objects, but it also comes with support for import/export facilities to/from HDF5/PyTables tables and pandas dataframes.

bcolz objects are compressed by default not only for reducing memory/disk storage, but also to improve I/O speed. The compression process is carried out internally by Blosc, a high-performance, multithreaded meta-compressor that is optimized for binary data (although it works with text data just fine too).

bcolz can also use numexpr internally (it does that by default if it detects numexpr installed) so as to accelerate many vector and query operations (although it can use pure NumPy for doing so too). numexpr can optimize the memory usage and use multithreading for doing the computations, so it is blazing fast. This, in combination with carray/ctable disk-based, compressed containers, can be used for performing out-of-core computations efficiently, but most importantly transparently.

Just to wet your appetite, here it is an example with real data, where bcolz is already fulfilling the promise of accelerating memory I/O by using compression:

http://nbviewer.ipython.org/github/Blosc/movielens-bench/blob/master/querying-ep14.ipynb

Rational

By using compression, you can deal with more data using the same amount of memory, which is very good on itself. But in case you are wondering about the price to pay in terms of performance, you should know that nowadays memory access is the most common bottleneck in many computational scenarios, and that CPUs spend most of its time waiting for data. Hence, having data compressed in memory can reduce the stress of the memory subsystem as well.

Furthermore, columnar means that the tabular datasets are stored column-wise order, and this turns out to offer better opportunities to improve compression ratio. This is because data tends to expose more similarity in elements that sit in the same column rather than those in the same row, so compressors generally do a much better job when data is aligned in such column-wise order.

So, the ultimate goal for bcolz is not only reducing the memory needs of large arrays/tables, but also making bcolz operations to go faster than using a traditional ndarray object from NumPy. That is already the case in some real-life scenarios (see the notebook above) but that will become pretty more noticeable in combination with forthcoming, faster CPUs integrating more cores and wider vector units.

Requisites

Python >= 2.6
NumPy >= 1.7
Cython >= 0.20 (just for compiling the beast)
Blosc >= 1.3.0 (optional, as the internal Blosc will be used by default)
unittest2 (optional, only in the case you are running Python 2.6)

Building

Assuming that you have the requisites and a C compiler installed, do:

$ pip install -U bcolz

or, if you have unpacked the tarball locally:

$ python setup.py build_ext --inplace

In case you have Blosc installed as an external library you can link with it (disregarding the included Blosc sources) in a couple of ways:

Using an environment variable:

$ BLOSC_DIR=/usr/local     (or "set BLOSC_DIR=\blosc" on Win)
$ export BLOSC_DIR         (not needed on Win)
$ python setup.py build_ext --inplace

Using a flag:

$ python setup.py build_ext --inplace --blosc=/usr/local

Testing

After compiling, you can quickly check that the package is sane by running:

$ PYTHONPATH=.   (or "set PYTHONPATH=." on Windows)
$ export PYTHONPATH    (not needed on Windows)
$ python -c"import bcolz; bcolz.test()"  # add `heavy=True` if desired