Yesterday I started playing with the HTML5 canvas as part of Google’s devart (art made with code) competition. Having never used the canvas before, I decided to start simple and build a sort of Jackson Pollock-esque random line generator.
This context is a built-in HTML5 object on which all the properties and methods for drawing on the canvas are defined. What is interesting to note here is that although the explicit call to “2D” in the getContext invocation would seem to imply that there are perhaps other kinds of contexts to the canvas (3D? 4D?!!), there are not. According to the canvas writeup on diveintoHTML5.info, while there are some unstandardized third party implementations of what a 3D canvas context might look like, the official HTML5 spec says that it does not exist yet.
Now that we have a context for our canvas, let’s write a couple functions utilizing its built-in methods to draw random lines on the canvas.
Briefly, this code randomizes the color, width, length, and position of a line, and then draws it on the canvas. Click the button below to see what 500 calls to drawLine() (with no lengths specified) produces on a blank canvas.
I think the effect is kind of beautiful. But ultimately, these are just random lines. There’s nothing really cool going on mathematically, and we have now laid the groundwork to inject some logic into our drawing process. This is where things get interesting.
It was about this point in my experimentation when a good friend (and Econ-Math major at Columbia, as will become clear shortly) peeked over my shoulder at what I was doing. He asked how all the lines were being generated, and I explained that I was randomly simulating their positions. Then, he told me about Buffon’s needle problem.
Buffon’s needle problem is one of the earliest and most famous geometric probability problems in mathematics. It goes like this: given a plane with evenly spaced vertical lines, we randomly drop needles onto the plane with the goal of finding the probability that a needle lands crossing a line. For simplicity, we assume that the length of the needles is equal to the distance between vertical lines.
As it turns out, Buffon’s experimentation led him to discover an early method to estimate pi, which, when holding the length of the needle and distance between vertical lines equal, comes out to two times the total number of needles dropped, divided by the number of needles that landed crossing one of the vertical lines.
This is a pretty easy problem to simulate. There are basically two things we need to do: generate lines of a uniform length (see the drawLine() function defined above), and determine if a line crosses some x coordinate. This is straightforward:
Given these methods we can write up a simulation and visualization of this problem for the canvas we were working on earlier. Though not very aesthetically pleasing, it actually gives pretty good estimates of pi! (Note that this is just running 500 trials, we would expect the number to come much closer to actual pi as we increase the number of trials.)
Initially, our simulations were yielding numbers of crosses that were too high, and thus estimates of pi that were too low. We couldn’t figure out why this would be for an hour or so, and decided to sleep on it. But as soon as my head hit the pillow, I realized that the problem lay in the implementation of the method getRandomX():
See the problem? We were flooring the result of our randomly generated x coordinates, making it much more likely that they would cross a vertical line, which are located at integer coordinates (multiples of 10 in this instance). Boy, that was dumb. When we corrected that issue, our numbers started to look a lot more normal.
Bringing it back
By this time it was 5:00am and my buddy and I were tired, so we turned in for the night. But I’ve since gotten to play around with the simulation we built and it yields some cool looking devart. Check it out!