MATLAB Programming Contest BlogApril 12th, 2006Tuesday Leap WinnerMarkus is the winner of our Tuesday Leap challenge. His entry, submitted at 9:31AM, made the single biggest improvement to the top score. Meanwhile, David Jones has been pushing the timeout and coming up with a best result of 82058 (85008 is the current leader’s result), showing there’s plenty of room for further improvement if you can find a way to compute it efficiently. The timing penalty grows exponentially, but the “secret” constants are tuned so that it is quite flat below a certain threshold. (Has anyone back-calculated these contents yet?) Of course, the leading entries tend to gravitatate to elbow. The “results vs. time tradeoff” plot at the bottom of the Statistics show this in action. 3 Responses to “Tuesday Leap Winner”Leave a Reply |
|
|||
|
These postings are the author's and don't necessarily represent the opinions of The MathWorks. |
Constants are:
k1 = 0.01
k2 = 0.001
k3 = 0.125
Interestingly, the ‘elbow’ is somewhere in the 75 - 95 second range, depending upon how you define it. The curve is flat until 60-65 seconds, which means the current leading entries with 60-61 seconds are right at the point where any additonal time gain has to be offset by result reductions.
At the current point in the curve with ~85000 as the result and ~61 seconds time, each point taken off the result equals ~0.05s of extra time allowed for computation.
Since the tweakers are steadily pushing the time down by tenths of a second, they are probably on the right path unless some major algorithm change occurs that drops a lot of points off the result. I suspect, unfortunately, the contest is going to come down to whomever gets a ‘lucky’ break in the timing variation between 5 and 10PM tonight since the queue will probably be spammed this afternoon.
I stand a bit corrected. Just as I was composing that last note, an algorithm improvement bumped the results down 677 points, while increasing the time by only 10s. Perhaps there is hope yet for the spirit of collaboration.