Have you encountered the following scenario?

You are trying to solve a problem (or helping solve a problem) and know or at least think you know the solution.  You are in the middle of implementing it when someone else looks at it and says, “why don’t you do it this way, isn’t this way easier/better?”  Taking a step back, you realize that the question not only has merit but is a better and much more obvious solution; you can’t believe you missed it.

What happened?

I think its because you were too close to the problem and had developed a very narrow focus.  That narrow focus prevented you from seeing the better solution.  Perhaps this is even a variation of functional fixedness in that we’ve latched onto an idea of how to solve a problem and our mind’s may not see alternatives easily.

What can we do?

  • Think about the broad (or product) level goals regularly.
  • Entertain questions and/or suggestions from others.
  • Ask: “Is this the best way?”
  • Ask: “Is this the practical way?”
  • Don’t overthink the problem.
  • Get it working then evaluate the solution and/or do a code review!

n_queens_graph
n_queens_table

Some History

This is a followup to my original post: Node.js vs Python vs PyPy – A Simple Performance Comparison.  This article corrects a discrepancy caused by a slight difference in the JavaScript implementation which skewed the Node.js results.

The Algorithm

As stated in the previous article, I’ve attempted to implement the same single-thread, brute force, recursive algorithm in many different languages.  There is nothing overly special about this algorithm and I’ve made no attempts to optimize it.

The Findings

Node is fast, very fast.  It easily outperforms any of the other implementations I’ve included in the puzzle’s repository.  As you can see by the included charts, the performance difference between Node.js and out-of-the-box Python is very significant and the difference between it and PyPy while less pronounced is significant.

Special Notes

  • I’ve placed my source on GitHub at the following url: https://github.com/chaddotson/puzzles.  It now contains functional N-Queens puzzle implementations in JavaScript, Python, Lua, Scala, and Ruby.  There is also a version in Rust, but that needs to be updated to the latest syntax before it can be run again.
  • This is just with one type of algorithm, the best solution might and probably does change depending on what type of application you are researching.  For webserver performance, Node.js is slightly better than PyPy running Tornado.
  • This algorithm is a simple brute force algorithm, there are many faster and better ones out there.
  • See the original article for the Node.js vs Python vs PyPy – A Simple Performance Comparison for more details memory performance.

Work is progressing on my first iOS app written in Swift.  I think that I might have a rudimentary alpha product ready to start testing in a few weeks.  Based on my experience with it so far, here are some of my thoughts, observations, and concerns.

  • Developing in Swift is as the name implies, “Swift.”  I didn’t know anything about Objective-C nor Swift before I started my app and I think I’m making decent progress given the amount if time I’ve had to work on it.  Given proper resources, I believe its possible to progress Swift apps from concept to production in a matter of weeks.
  • Language resources (documentation and examples) are still pretty sparse.  Get ready to learn a little Objective-C if you need help with API calls because that is where the majority of the help you will find will be.
  • The iOS 8 Simulator doesn’t save location privacy settings,  That means that every time you run an app that requires location privileges, you have to edit the settings again.  Based on what I’ve read, this issue has existed for some time in the beta code.
    This was due to user error.  You have to use either the location manager’s requestWhenInUseAuthorization or requestAlwaysAuthorization functions paired with the right entries in your projects Info.plist file.  The reason mine wasn’t working was I had erroneously edited the plist file associated with the unit tests.
  • Perhaps the biggest issue is that Xcode cannot refactor Swift code.  To me this seems like a giant shortcoming of the editor.  Hopefully Apple will push an update soon to correct this.
  • While storyboards are powerful, some of the functionality is not so discoverable.  The auto-layout functionality took me awhile to find and not without searching the net.  I think they maybe onto something, but I do tend to like the way Visual Studio makes similar functionality “findable” from the properties panel.
  • As of Xcode 6.0.1 and iOS Simulator 8.0, AVSpeechSynthesizer doesn’t appear to work at all.

    When executed in the iOS simulator, the code results in “Speech initialization error: 2147483665”.  I’ve seen a few work arounds on the net, but I don’t believe any of them actually work in Swift.

Swift and Xcode 6

Late last week Apple released Xcode 6 to the general public.  I’ve been waiting for the opportunity to try out Swift, so I started working on an idea I’ve had.

General Thoughts

This is probably a general iOS development comment, but the way you link items on the storyboard to class members seems odd to me (having done a lot of C# development in the past).  However, it does have its advantages.  Counter to the Microsoft environment, it does seem to encourage better design and discourage the worse of some bad habits.

Compared to Android

Its been awhile since I worked with Android, but when I did developing for the platform was disastrous.  Setup of the environment (Eclipse) and the emulator was time-consuming and not so straight-forward.  With Xcode, its all on rails.  You can have a “Hello World” app up in the simulator in 2 minutes.

Problems

The language does seem to be a little wordy at first glance maybe that opinion will get better as I get more experience with it.  While the official Apple documentation is now complete with Swift equivalents to Objective-C,  examples both official and third party are pretty few and far between it seems.

To be continued…

Picture of my simple led circuit connected to the Pi.

Picture of my simple led circuit connected to the Pi.

Programming on the Raspberry Pi

Python seems to be the more popular language for writing programs on the Raspberry Pi.  However, it is far from the only language: Python, C, C++, Java, and Ruby are some that are automatically supported out of the box.  As it turns out Node.js is also supported.  Perhaps, just based on my preconceptions, I didn’t originally consider JavaScript when interfacing directly with hardware.  The goal of this article is to compare functionally equivalent sample programs written both in Python and JavaScript.

 

The Goal

I decided to make the scope of the problem created for this article as small as possible.  Given that the goal will be to create simple circuit diagram consisting of a LED and then to write a program that makes the LED flash for a specified period of time.

 

Simple LED circuit diagram

Simple LED circuit diagram

Hardware

With the Pi, a circuit that will make a LED flash is relatively simple: Consisting of just a LED and resistor.  I arbitrarily selected pin 7 on the Pi for this build.  I’ve included a diagram of it on this page.

 

Software

As the goal states, the following two implementations do nothing except toggle an LED at half the specified duration (0.5s) and automatically shutoff after 60 seconds.

 

Python

Since Python is the more traditional programming language for the Raspberry Pi, let us start with it.  The Rpi.GPIO module I used can be installed via pip.  Before you read the code sample, let me point out some things about the implementation.  Could it have been done simpler given that the problem was just to make the LED flash for a certain amount of time?  Yes, but I wanted to create something of an asynchronous process that could be controlled from outside its execution context.

 

JavaScript

I used pi-gpio to interface with the GPIO on the Pi in JavaScript.  It can be installed via npm. It makes use of gpio-admin so that your script doesn’t have to run as sudo.  Follow the installation instructions provided here and it should get you setup.  Note that due to differences int the languages, it was not necessary to thread (web workers) the solution.

 

The Findings – I was surprised

When I started writing the JavaScript side of this article, I was mainly doing it to gain experience working with JavaScript on the Pi.  I did not expect to come out of it vastly preferring it over my python implementation.  I find the Python implementation above to be to wordy and overcomplicated.  It just seems that the amount of code needed to achieve the same results is excessive.  Perhaps this is because of the problem scope and implementation.  The problem posed here was a simple one, where a functional solution is superior to an object-oriented one.  Could the Python code above be rewritten into something a bit more functional? Sure!  Are there problems that an object-oriented Python or JavaScript implementation would be the better solution for?  Definitely.  The lesson to take away: be open to working outside your comfort zone and pick solutions to problems based on their fit for the problem not your comfort level with them.

I’ve created a copy of the code as a Gist that is available here: https://gist.github.com/chaddotson/570501a3e3dcfe8928c8

NQueensGraph

IMPORTANT NOTE: The NodeJS algorithm had a slight discrepancy in it.  See this article for a correction to the performance comparison section of this article.

The Algorithm

Yesterday, I decided to try translate my algorithm for calculating N-Queens to JavaScript.  I’ve implemented the same single-thread, brute force, recursive algorithm in many different languages with the biggest difference being the syntax of the language.  Once I completed the JavaScript Implementation, I ran the program with the latest version of Node.js.

The Findings

I knew Node was fast but it still surprised me.  As you can see by the included charts, the performance difference between Node.js and out-of-the-box Python is pretty significant.  Its not until the algorithms complexity and recursion depth hit certain limits that Node.js’s performance starts to falter.

Node.js and CPython – What’s The Difference?

You might ask what is behind this performance difference.  The answer is actually pretty simple.  It all boils down to how the code is being executed.  Node.js uses the V8 JavaScript Engine (Wikipedia | Google) written by Google and a part of the Chrome Browser.  V8 includes a just-in-time compiler that compiles the JavaScript to machine code before execution and then continuously optimizes the compiled code.  Python is a bytecode interpreter; meaning that the default interpreter (CPython) doesn’t execute Python scripts directly.  Instead, it first generates a intermediate file that will later be interpreted at runtime.

Ways To Get Better Performance

If you want to use Python, we can overcome the differences between Node.js and vanilla Python by using PyPy, an alternative implementation of Python that includes a just-in-time compiler.  For the algorithm I wrote, you can see a pretty good performance boost over Node.js when using PyPy.

Special Notes

  • I’ve placed my source on GitHub at the following url: https://github.com/chaddotson/puzzles
  • This is just with one type of algorithm, the best solution might and probably does change depending on what type of application you are researching.  For webserver performance, Node.js is slightly better than PyPy running Tornado.
  • This algorithm is a simple brute force algorithm, there are many faster and better ones out there.
  • At a board size of 15, Node.js could no longer run the algorithm due to its maximum recursion limit.
memory_usage
memory_usage_chart

Edit – A Follow-Up

The original focus of this article was shear performance, but I’ve received a question regarding the memory footprint of the 3 methods.  I think that is a very good and valid question.  So, I reran the tests to capture the peak memory utilization by each.  For this test I used “/usr/bin/time -l” to capture the maximum resident set size.  While this isn’t exactly the peak amount of memory utilized by the algorithm, it is sufficiently close to report on.

New Findings

Upon rerunning the tests for capturing memory utilization, I found that for the most part memory utilization contrasts performance.  A higher memory utilization isn’t really unexpected, if you think about it.  Essentially, the jit is sacrificing memory for performance.  In most cases, this isn’t really that bad.  Using a jit is just a cheap way of boosting performance of code written in an interpreted language.  The boost in performance, speed of which it was written and the maintainability of it outweigh memory utilization concerns in many cases.

The Oddity

As you can see, I’ve included a chart covering all the solutions for boards 8×8 to 14×14.  During most increments in board size, the memory utilization seems to increase exponentially; however, when we hit the 14×14 board size we see all the cases level off at relatively the same memory utilization of around 300 MB.  At this time, I really don’t have a good answer for this.  I could certainly speculate, but I’d rather not until I know more.

 

“Measuring programming progress by lines of code is like measuring aircraft building progress by weight.” – Bill Gates

I have that quote on my instant messenger. I found out Friday that someone had saw and used it. Its good that someone found it useful. It’s a very valid point to make that lines of code isn’t an accurate measure of programming progress nor complexity. As a place to start, people looking for better methods might like Feature-driven development with a meaningful estimating system like PERT, CPM, or other system.