Charting Reaction Times on the Raspberry Pi

Earlier this week I collected some reaction timer data on my Pi using the BBC micro:bit as an input device. I only produced an “ASCII art” chart at the time:

      times←ReactionTimer.Play
      times
251 305 294 415 338 298 294 251 378
      ReactionTimer.AsciiChart times
425| 
400|    * 
375|         *
350+ 
325|     * 
300|  * 
275|   *  ** 
250+ *      * 

Retro is back in style – but of course I should point out that we can produce “proper” graphics using SharpPlot, a cross-platform graphics package that is included with Dyalog APL on all platforms. I’ve enhanced the ReactionTimer namespace with a function called SPHistogram. If you are using RIDE as the front end to APL on the Pi, this will render output from SharpPlot in a window:

    ReactionTimer.SPHistogram times

Reaction Time Histogram

The original ASCII chart simply plotted the observations in the order that they occurred. Above, the shaded areas show how many observations there were in each bucket (2 between 250ms and 275ms, 3 between 275ms and 300ms, and so on). At the same time, the individual observations can be seen along the X axis as vertical red lines. It is a little unfortunate that, presumably due to some artifact of the timing process, the values 251 and 294 occur twice – and these lines are drawn on top of each other.

The example highlights one of the things that makes SharpPlot a bit special: We have overlaid a histogram showing the frequency of reactions in each bucket, and used a “scatterplot” to where the Y value is always 0, to mark the individual observations along the X axis.

     ∇ SPHistogram times;heading;renderHtml;sp;svg;z 
[1]   heading←'Reaction Times' 
[2]   renderHtml←3500⌶ ⍝ Render HTML in Window 
[3] 
[4]   :If 0=⎕NC'#.SharpPlot' ⋄ #.⎕CY'sharpplot.dws' ⋄ :EndIf
[5] 
[6]   sp←⎕NEW #.SharpPlot(432 250) 
[7]   sp.Heading←heading 
[8] 
[9]   ⍝ Draw histogram 
[10]  sp.ClassInterval←25 
[11]  sp.SetXTickMarks 25 
[12]  sp.HistogramStyle←#.HistogramStyles.SurfaceShading 
[13]  sp.SetFillStyles #.FillStyle.Opacity30 
[14]  sp.DrawHistogram⊂times 
[15] 
[16]  ⍝ Add observations using ScatterPlot 
[17]  sp.SetMarkers #.Marker.UpTick 
[18]  sp.SetPenWidths 1 
[19]  sp.SetMarkerScales 3 
[20]  sp.DrawScatterPlot(times×0)(times) 
[21] 
[22]  ⍝ Render SVG and display in window 
[23]  svg←sp.RenderSvg #.SvgMode.FixedAspect 
[24]  z←heading renderHtml svg 
     ∇ 

Hopefully the code is more or less self-explanatory, but if you’d like to learn more about SharpPlot there is excellent documentation at http://sharpplot.com. The documentation is actually written for C# users, but there is an illustration of how to translate the documentation to APL (and VB) at http://www.sharpplot.com/Languages.htm.

micro:bit Reaction Timer in APL on the Pi and BBC micro:bit

BBC micro:bit displaying a happy face

BBC micro:bit displaying a happy face

I have a bit of a cold today, so I decided that instead of hopping in an icy car and driving to the office in order to spend the day drinking coffee and answering e-mail, I should stay at home, turn up the radiators, make lots of tea (with honey!) and have some fun writing code on my Raspberry Pi! Can there be a better way to ensure a speedy recovery?

By the way, if you are already a user of Dyalog APL but you haven’t got a Pi yet, you should read The APLer’s Quick-start Guide to the Raspberry Pi, which Romilly Cocking completed a few days ago. This explains what your options are for buying hardware, and how to get going with a free copy of Dyalog APL. If you don’t read it now, your cold may be over before all the bits are delivered, so hurry up!

I wasn’t feeling THAT energetic this morning, so I decided to ease back into things by trying to replicate Romilly’s reaction timer, which is written in MicroPython. To begin with, I extended the microbit class that I developed for the Morse code display function with a few new methods to cover the API calls that allow me to check the state of the two buttons on each side of the micro:bit display (see the image above).

First, I added a method called is_true, which takes a MicroPython expression as an argument, evaluates it using the PyREPL function, and returns 0 or 1 depending on whether the expression returns False or True (and fails if it returns something else):

 ∇ r←is_true expr;t;z
 :Access Public
 r←1⊃t←'True' 'False'∊⊂z←PyREPL expr
 :If ~∨/t ⋄ ⎕←'True/False expected, got: ',z ⋄ ∘∘∘ ⋄ :EndIf
 ∇

This is a useful building block, which allows the simple construction of functions like was_pressed, which takes ‘a’ or ‘b’ as an argument and tells you whether the button in question has been pressed since you last asked:

∇ r←was_pressed button
 :Access Public
 r←is_true 'button_',button,'.was_pressed()' 
∇

Once this is done, writing the Play function in the ReactionTimer namespace is easy, you can find the rest of the code on GitHub. The APL version is different from the MicroPython version in that – once the user presses button B and stops the game – the reaction times (in milliseconds) are returned as an integer vector. So now we can have some fun with data!

In the spirit of the micro:bit, I thought I’d produce a low tech character based chart, trying to remember the skills that were start-of-the-art when I started using APL. The AsciiChart function in the ReactionTimer namespace takes a vector of timings as the right argument, and produces a chart:

      times←ReactionTimer.Play
      times
251 305 294 415 338 298 294 251 378
      ReactionTimer.AsciiChart times
425| 
400|    * 
375|         *
350+ 
325|     * 
300|  * 
275|   *  ** 
250+ *      * 

The code (which is also on GitHub) is listed below. Because this is a pure function, I used the modern dfns style of function definition rather than the old style procedural form that I’ve used for the object oriented code. The function works by allocating each value to a bucket of the size defined by the variable scale. The fun part is the use of the outer product with equals (∘.=) between the list of buckets on the left (rb) and the scaled values on the right (⌊⍵÷scale) – and then using this Boolean array to index into a two-element character vector to produce the chart. The rest is scaling calculations and finally decorating with “tik marks”:

 AsciiChart←{
     scale←25 ⍝ size of each row
     tiks←4 ⍝ tik spacing
     (max min)←(⌈/ , ⌊/) ⍵ ⍝ maximum and minimum
     base←⌊min÷scale ⍝ round down to nearest scale unit
     rb←base+0,⍳⌈(max-min)÷scale ⍝ row base values
     r←' *'[1+rb∘.=⌊⍵÷scale] ⍝ our chart
     r←((≢rb)⍴'+',(tiks-1)⍴'|'),' ',r ⍝ add tiks
     r←(⍕⍪rb×scale),r ⍝ add base values
     ⊖r ⍝ low values last (humans!)
 }

Although I might be feeling better tomorrow, I expect to be back soon. I have a couple of plastic bags full of resistors, LEDs, potentiometers and cameras (yes, plural) that I have not opened yet!

Morse Code – Revisited using the BBC micro:bit

As mentioned last week, I have found a new way to provide a front end processor for my Pi, the BBC micro:bit. This week, I have started putting the new system through its paces: The microbit class in our GitHub repository has been beefed up to make it more resilient in the case where serial I/O is not initialised, or the MicroPython REPL is not running, and I have added a trivial Python program which can be used to intialise the REPL using the mu editor.

To test it all, I have revisited the Morse Code example, originally written to use GPIO support provided by Quick2Wire, to see if I could get it up and running with my new infrastructure. The micro:bit has a 5×5 LED display which makes the result a bit more impressive to look at:

 

The new code is significantly simpler than the original, for a couple of reasons:

  1. Making calls to the micro:bit API via MicroPython is easier than using GPIO via a C library.
  2. Version 15.0 of Dyalog APL contains new system functions for working with text files, so we no longer need the “Files” library, we can just use the built-in ⎕NGET function to read the contents of a file.

Raspberry APL Pi and Python on the micro:bit

A couple of years ago I spent many happy hours writing APL code to control robots which each embedded a Raspberry Pi. It was fun but it was a bumpy ride – my interest eventually faded when I discovered that it was just too difficult to make sense of raw accelerometer data, which I was hoping to use for precise navigation. After I managed to fry my MPU-9150, I decided to pursue other interests, as they say.

Recently, Romilly Cocking has been waxing lyrical about a number of new components which he thinks will make the whole experience more enjoyable – and I have decided to dive back in. So far, it has indeed been a very pleasant experience.

The Raspberry Pi Model 3

The Raspbery Pi Model 3While I was away, the Raspberry Pi and the software available for it have come a VERY long way. The speed of a Pi 3 makes it a perfectly usable computer. I am writing this blog entry on the Pi using a standard Chromium browser. APL starts up immediately, using our new graphical IDE. It is quite simply amazing!

The BBC micro:bit and MicroPython

The first generation of APL-based robots used an Arduino to manage the motors and sensors because timing was too critical for the Pi to handle it. I spent some rather unhappy hours writing a control program in C, debugging that without any tools. You could not even do “printf debugging” to the serial port because that interfered with the timing so much that it stopped everything from working.

Romilly suggested I take a look at the micro:bit, and since I have been using an interactive REPL all my life, he suggested that I run MicroPython on it.

He was right, I’m loving it!

Getting Started

I purchased the Kitronik Inventors’s Kit for the micro:bit from Pimoroni, and quickly found videos on Youtube explaining how to assemble it. I installed the “mu” editor on my Pi and was immediately able to edit a simple Python program and run it on the micro:bit.

My next step was, of course, to create a little class to drive MicroPython from APL. This involves using serial I/O: when the micro:bit is attached to your Pi via USB it appears as a folder and as a serial (tty) device, typically /dev/ttyACM0. You can find my code in the APLPi/microbit GitHub repository.

If you clone the repository to /home/pi/microbit and start APL, you can make calls to the MicroPython repository along the lines of:

    ]load /home/pi/microbit/microbit    
    mb←⎕NEW microbit ''
    mb.PyREPL '2+2'
4

Using Python on the micro:bit to do maths isn’t that interesting when you have APL on the Pi; the exciting thing is that we can call the micro:bit Python API and access the hardware. My first experiment was to display my favourite APL symbol (Grade Up – aka Imperial Cruiser) in the built-in 5×5 LED display:

Imperial Cruiser on  the micro:bit

APL displaying an Imperial Cruiser on the micro:bit

Which of the Inventor’s Kit experiments would you like to see me do next?

The 2016 Year Game

Our 2016 Year Game was launched in January 2016 and ran until the end of the year. The idea was simple – to find APL expressions involving exactly the digits 2 0 1 6 in that order to equal the numbers 0 to 100 using the fewest characters possible. The minimum number of characters for an expression is 5 (the four digits 2 0 1 6 and a single primitive function/operator) so the smallest number of characters that a possible solution can have is 505.

A lovely elegant solution was pointed out by Lars Stampe Villadsen:

      ⎕IO←0
      {⍪⍳+/⍵*⌽⍵}2 0 1 6

and Bernard Legrand submitted two one-line solutions:

      (⍴⍬),⍳¯20+!1-⍨6

and

      (⍴⍬),⍳⍴'Be happy: this solution to Game 2016 will be the most stupid one, but it needs only five APL symbols'

These three each return all the numbers from 0 to 100. Not valid submissions for this particular game, but I did find them rather pleasing. Anthony Cipriano was inspired to go one step further and submitted an expression that creates the first 8 Fibonacci numbers:

      (⊢,(+/{(¯2∘↑)⍵}))/(⊂0 1),⍨⍳6

The wonderfully varied ways in which APL can be used (and in which APL users think!) was illustrated beautifully by some of the solutions. Although some numbers were attained in the same way by everyone (36, for example, was always 20+16), others lent themselves to multiple solutions. No-one managed to calculate 100 in fewer than 7 characters, but the 7-character answers included 20∧⌹.16, 20ׯ1+6, -20×1-6, 20×-1-6, ¯20×1-6 and various others on the same theme. Unity experienced similar creativity, with 1 being reached in the minimum possible number of characters (5) using ×2016, 201≢6, 20>16, 20≠16, 2<016, etc.

It was clear early on that submissions fell into two distinct groups – those who wrote a program to find the expressions and those who derived their expressions manually. I found the creativity evident in the manually-derived solutions very satisfying; the results may not have been as terse as those resulting from programs, but the individuality of the creators shows people inspired by their subject matter. Who can fail to appreciate the thought process behind arriving at 96 using (×/+\+\(!2 0 1))+6 rather than the more prosaic ⌈⍟!20+16 or ⎕IO-⍨+⍨20+16 rather than 20+⌈○16 for 71?

Having said that, the goal was conciseness and individual character counts are now tabulated on the 2016 Year Game page; an amalgamation of the lowest-character expressions that anyone achieved can also be downloaded from this page. Congratulations to Jonas Stensiö and Veli-Matti Jantunen, who both achieved the same character count as the amalgamated minimum (693).

(most names omitted to protect the not-so-innocent)

Dan Baronet

Dan Baronet

Dan Baronet

We do not know the details, but Dan’s luck ran out this week. On Tuesday, as he headed South from Nevada into California, en route to Mexico on the first leg of a four month motor cycle adventure to South America, something went wrong … Dan lost control of the bike and collided with a motor home, and many of us are now struggling to comprehend that Dan won’t be back.

Dan made friends everywhere that he went. On our annual Dyalog skiing “retreat” (we frequently did real work on prototypes of new tools, honest!), I was always impressed (while trying to hide my slight embarrassment) by the way Dan always kicked off conversations with hotel staff. Whether the waiters and waitresses were native Italians from the resort town – or migrants as hotel staff frequently are – Dan would typically know enough words in some shared language to get the conversation off to a good start (or he would extend his vocabulary on the spot). This always guaranteed us service with a smile, and excellent advice on which wine to order for dinner.

Dan gave as well as he got: on the slopes, Dan was always the first to stop by anyone who had fallen over and was struggling to get back up. Dan has friends all over the world and always took the time to visit them when he could, to keep the friendships alive. He was also happy to open his own home to travellers, both my kids had little Canadian adventures as guests of Dan and his family in Quebec, and Dyalog team members would also “couch-surf” chez Dan.

At work, Dan also loved extending his vocabulary. He remembered more details of different APL dialects than anyone else that I know, and was an expert in migrating code from one system to another. He also knew the weaknesses of each system, and was a feared visitor to vendor booths at APL conferences. Dan would mischievously sit down to play with a new APL system, try a couple of his favourite edge cases, and usually manage to crash new (and some old) APL interpreters within seconds.

But even when crashing interpreters, Dan was really trying to help the vendor produce a more stable platform. At work, as on the ski slopes, Dan was one of the few people who was always going out of his way to help. When he had spent time learning a new technique, or how to use a new tool, he would go the extra mile to create a utility library, document it, presented it at conferences and APL user meetings. He also did the really hard part – writing an article about it, making frequent contributions to Vector and other publications: http://archive.vector.org.uk/?qry=baronet&submit=search. He also enjoyed teaching APL, as an instructor at I.P.Sharp Associates and in the modern era as the author of numerous tutorials at https://www.youtube.com/user/APLtrainer.

More than 30 years ago, as an APL rookie at I.P.Sharp Associates, I was fortunate to meet, work with, and learn from Dan when he and I both found ourselves at IPSA Copenhagen. In addition to a general interest in tools, I shared Dan’s interest in the migration of code between APL systems, and flying airplanes. Dan actually flew small aircraft (and helicopters!) and I, like so many other friends and colleagues, admired him and got invited along for rides.

A decade and a half later in 2005, as the new CTO of Dyalog, I was keen to add an APL toolsmith to the company. Despite his location in Montreal, Dan was at the top of my list. Dan continued to live in Montreal but was a frequent guest in Bramley, coming over several times a year to spend some weeks to work with the growing “APL Tools Team” – and with the core development team on improving the quality of the interpreter by crashing it in as many ways as he could think of.

When his imagination ran out, he wrote code to generate even more ways to crash the interpreter. Dan has been cursed loudly many times, and will be remembered – if for no other reason than that it will be a long time before we are able to close the last “issues” that he logged in our problem tracking system. The quality of Dyalog APL has increased enormously in the last decade, and we have much to thank Dan for. Dan also “poured the concrete” for the foundations of the source code management tool (SALT) and the User Command framework that we are all starting to take for granted, and contributed to many other pieces of our growing collection of tools.

C-GSXN in Toronto

Dan often flew himself to APL User Meetings

Dan was an adventurer. He skied, snowboarded, snowshoe’d, biked, skated, hiked and flew – and with his great sense of humour and warm personality, he was able to entice many of us to come along with him to share. He photographed and brought his video camera, although he would often manage a 3-week trip to Europe which included a skiing trip, carrying only a small backpack.

Dan had just turned 60, and he had taken four months leave this winter to pursue an old dream to revisit South America with a friend – on motor cycles. They had wisely gone on a trial ride in the late summer, riding from Canada to Las Vegas and parking the bikes there in order to test the equipment and not have to ride two-wheelers in the Canadian winter. As a pilot, Dan understood risks very well – or rather, how not to take unnecessary ones.

“If only he had been a little more cautious”, some will say. Was that trip really necessary? These thoughts are unavoidable – but we have to remember that his decision to ride was natural to the adventurous and mischievous spirit that we all knew and loved.