Wednesday, August 24, 2016

The Julia programming language: amazingly nice

Well, at least I am amazed. I took a brief look at Julia a few years ago but since I understood it to be somewhat derivative of GNU/Octave (or Matlab) and R, and since even though I use GNU/Octave and R I don't enjoy these languages, I only gave Julia a very short look.

Fortunately, a current customer uses Julia so I have been ramping up on the language and I very much like it. A bit off topic but I would like to give a shout-out to the O'Relly Safari Books Online service which I recently joined when they had a $200/year guaranteed for life subscription price (half regular price). I am reading "Getting Started with Julia" by Ivo Balbaert which is fine for now. I have "Julia for Data Science" by Zacharias Voulgaris and "Mastering Julia" by Malcolm Sherrington in my reading queue. When learning a new technology having up to date books available really is better than finding information on the web (or at least augmentation to material on the web).

I very much like the tooling for Julia. Julia is a new language but there are already many useful libraries available. Julia uses github for storing modules in the standard library and the integration works very well, at least on Ubuntu Linux. So far, I have been happy just using GNU/gedit for development. I haven't tried Julia on OS X or Windows 10.

The Julia repl is great! Color coding and auto completion are especially well done.

I like just about everything about the Julia language except for 1-based indexing of matrices. Oh well.

Julia is readable, functions are first class objects and programming in Julia is very "Lisp like." With optional type hints (mostly in defining function arguments) Julia is a very high performance language. I love developing in Ruby but I do dream of much higher performance. Julia does not seem like a complete replacement for Ruby (for me) however. That might change.

In addition to doing work with Julia, I have also been experimenting with lots of little coding projects: the Merly web framework (simple, sort of like Sinatra), using the standard HiddenMarkovModels library, and experimenting with a few of the neural network libraries. All good stuff.

Sunday, August 21, 2016

My prediction: Immersive real-time VR in Olympic closing ceremonies in 8 years

My wife and I are watching the closing ceremonies right now. Great visual effects that will be even better with immersive virtual reality. I expect that in 8 years we will have the option of being able to change our point of view from the stands to down on the central floor in a complete immersive VR experience with 3D sound and head tracking.

I haven't worked in VR in almost 20 years when I helped found the virtual reality systems division at SAIC (where I handled 3D sound with head related transfer functions, motion, haptics, and some graphics) and then a year later did a virtual reality project for Disney while working at Angel Studios. Even if I don't work in VR anymore I am a huge fan and I have high expectations for what is to come in user experience.

Sunday, August 14, 2016

I was surprised that so many of the NACL 2016 papers described deep learning projects

I attended the North American Chapter of the Association for Computational Linguistics conference last June in San Diego. Here is a link to the published papers.

The conference was great. The keynote talks, panel discussions, and the talks I attended were interesting! As an independent consultant I payed my own way to the conference and I found it to be a good investment. Sometime I would enjoy attending a European chapter of the ACL conference.

Saturday, August 13, 2016

Some new love for Scala and Python

I am a practical developer. I do have my favorite programming languages (Ruby, Haskell, Clojure, and Java) but I tend to look first at what libraries are available in different languages for whatever project I am currently working on.

I did a lot of work in machine learning in the 1980s (mostly in neural networks) and since then I have probably spent about 15% of my work time directly working on machine learning problems. That has changed in the last few years since several of my consulting customers wanted help spinning up on machine learning.

I have used Scala a fair amount but it has never been a "favorite language," mostly because I didn't care for the tooling. Now I find myself motivated to use Scala because of the awesome Apache MLlib and Breeze machine learning libraries. Also, I have solved my "tooling problem" for Scala development; if you are interested here is my setup: I use a remote high-memory, high-CPU server instance for fast builds. I used to use IntelliJ for Scala development but now I just keep a SBT console open and use Emacs with Ensime and sbt-mode using SSH shells. This is a simple setup but now I am happier using Scala.

I have also been spending a fair amount of time with Google's TensorFlow deep learning tools and the easiest path to solving problems with TensorFlow is working in Python. If you are interested, I do almost all of my work with Python using the free community edition of PyCharm.

So, in general I am trying to avoid the "want to use my favorite programming language trap." The joy is in solving problems and not in wanting to use a favorite language and software stack.