The data science community received a special delivery in December with the release of Python 3.6. At the time, I had a conversation with The New Stack to discuss what’s new with this release, and why we at Continuum see 2017 as the year that Python 3 is beginning to dominate the data science landscape, with major adoption in the enterprise space. I’m excited that today we’re extending that story with the most recent version 4.4 release of Anaconda, available for both Python 2.7 and Python 3.6.
The data science community received a special delivery in December with the release of Python 3.6. At the time, I had a conversation with The New Stack to discuss what’s new with this release, and why we at Continuum see 2017 as the year that Python 3 is beginning to dominate the data science landscape, with major adoption in the enterprise space. I’m excited that today we’re extending that story with the most recent version 4.4 release of Anaconda, available for both Python 2.7 and Python 3.6. Anaconda downloads have skyrocketed in the past six months, with more than a million downloads per month. Besides delivering a comprehensive platform for Python-centric data science with a single-click installer for Windows, Mac, Linux, and Power8, Anaconda 4.4 is also designed to make it easy to work with both Python 2 and Python 3 code. If you haven’t already started using Python 3, there’s no better time than today.
Why Python 3.6?
When Python 3.6 launched, I was excited to see this tweet from legendary Python core developer and former Python Software Foundation board member, Raymond Hettinger, as it expressed my sentiments exactly:
To backup a minute; in 2016, Python 3.5 proved itself among the established Python community, and word got out that Python 3 was good to go. This made Python 3.6 the first version of the language that was delivered to a maturing base of users, poising it for a prime growth opportunity. Having just returned from PyCon 2017 in Portland Oregon—the flagship event for the Python community—I can tell you that Python 2 is mostly a footnote, and the buzz is largely around Python 3. With just over three years until Python 2 support is officially eliminated by the Python Software Foundation, and the scientific Python community publishing coordinated timelines for when Python 2 support will be suspended for many popular libraries, it is reasonable that there’s a widespread move to Python 3 happening now. This is great news because Python 3, and especially Python 3.6 (as Raymond alluded to in his tweet), offers many great new features and performance enhancements.
Of the 200 tools and libraries that Anaconda provides, there are dozens designed to help with co-development of Python 2 and Python 3 code. This means that many of the new Python 3 standard library capabilities are available through backported implementations that are bundled in Anaconda for Python 2. Enterprises that are still committed to maintaining their legacy Python 2 code-base can benefit from these features in advance of migrating to Python 3. Furthermore, Anaconda’s software sandboxing system means it is easy to run Python 2 and Python 3 in tandem on the same system, supporting a gradual migration and avoiding the need for any “big bang” cutover. For many, these capabilities alone are enough of a reason to use Anaconda. If you’re looking to migrate people and software to Python 3, then I’d recommend Lennart Regebro’s Python 3 Porting website and the shorter PSF HOWTO on Porting to Python 3 by Microsoft engineer Brett Cannon.
Anaconda 4.4 also ships with the Intel Math Kernel Libraries, providing a substantial performance boost for the 20+ Python libraries that are compiled to leverage these optimized routines on several generations of Intel processors. Numpy, Scipy and Scikit-Learn and all the libraries that build on these benefit from this.
We’re advising our clients to see Python 3.6 as a “reference release” that should be adopted for any new Python-based projects.
Python 3.6 offers a more stable version of the language that enhances some core concurrency capabilities around a concept known as “coroutines.” These provide new language constructs for the creation of asynchronous, and possibly parallel, functions. This is an area that is obviously important in the world of multi-core CPUs, increasing demands for computational power, and the leveling off of peak processor speeds. Python 3 also offers fundamental improvements of the core data structures, in particular dictionaries on which the language is practically built. Python dictionaries are often known as “hash-maps” in other languages.
The piece I’m most excited about, however, is the introduction of type annotations that can be used to provide a degree of type checking. I believe this will take Python to a new level for enterprise adoption and introduce possibilities for interface design and program behavior that have been tricky until now. At Continuum, we’re looking forward to leveraging these in our Numba Just-In-Time compiler for Python.
Join the Conversation
If you’re not already using Anaconda I’d encourage you to download it today. If you’re already an Anaconda user then you can update to the latest version with:
conda update conda anaconda
If you’re using Anaconda Navigator, update the “anaconda” package. If you want to get started with Python 3, then I’d recommend an O’Reilly booklet written by a colleague of mine, David Mertz, that describes the benefits and how to migrate to Python 3. It is called Picking a Python Version: A Manifesto.
I’d love to hear your thoughts and comments on Python 3 and Anaconda. Catch me on Twitter: @ijstokes.