 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|
|
|||||||||
| |||||||||||
|
|||||||||||
| |||||||||||
|
|||||||||||
|
Finally, dplass talks about Design and Development
05.03.06 7:30 PM  By dplass TopCoder Member This morning, during the Algorithm Room 1 round, I got to talking with some of the developers and designers who were here. I finally met the infamous visualage, who has a reputation for being a very fast coder. Yesterday, one of the TopCoder admins said that he once saw him code an entire component in 3 hours in a hotel lobby. That's fast. It usually takes me 30-40 hours (working mostly from 9 pm to midnight) for a component. So, I got to meet the man behind the legend. I made a joke, "Is doing TopCoder components your new profession?," which was pretty funny; I did once consider that, but the problem is that the income isn't guaranteed! We were able to turn that into a Div 1 500 dynamic programming problem - computing the "expected income" based on the relative difficulty of the component, probability of winning, and actual payment of the component. (This was all in jest, natch.) He was very gracious, saying that it was disappointing that I was not able to make the TCO '05 finals in Santa Clara. (I had personal conflicts with both the on-line finals and the on-site finals as it turns out. I believe cnettel progressed to the finals instead. He is here, but I haven't spoken to him yet -- despite eating lunch next to him.) Anyway, during the Algorithm round, AleaActaEst came over and introduced himself. It's his first on-site as well, so we have a similar view on the whole experience. We actually started talking about algorithm SRMs and how stressful they can be sometimes. "When I can't get a problem, I just feel my blood pressure rising," he said. I also used to be like that, but finally learned how to relax, even though you wouldn't know it by my rating. traugust came over as well, and the three of us chatted about the relative plusses and minuses of development and design competitions at TopCoder. The consensus was that it takes less time for design projects than development projects, but the trouble that I myself have with them is that there is too much left for interpretation in design. As a result, reviewers sometimes force "scope creep" by their own interpretation of the requirements, which may be different than the designer's view. Just then the proceedings were interrupted with: "Attention please! There are 2 minutes left in the coding phase...phase...phase"OK, now that scared the bejebus out of nearly everyone in the room! ntrefz's computer-bass-enhanced voice was a bit, er, LOUD.After the room had a nice laugh, the three of us continued. I feel that development is more "cut and dry" in this area - either you follow the design or you don't. But, occasionally you run into platform-specific problems. traugust related to us a problem he had with the Java "default locale", which of course is different in his native Germany than in China, where the reviewer was. Similarly, AleaActaEst had a development bug whereby all three reviewers, and the developer, were getting different results on different operating systems and JVM versions. Myself, while doing the final final final fixes (ivern might remember this as the final fixes that lasted 2 months) for the API Analysis component, I found myself having to work around a known bug in Java's URLClassLoader class. How frustrating! After lunch, the Design and Development competition started with similar fanfare to the Algorithm Room 1 Competition this morning. cnettel and traugust had their own cheering section, perhaps related to their nationality or ???. Today, the designers were appealing their "Relations" component, and the developers were appealing the "Document Indexing Persistence" submissions. All components had been submitted a month ago, which added to the challenge for the competitors. I was a bit ambivalent as to whether or not watching the competitors write appeals would make for good spectating, but went into it with an open mind. Similar to this morning, I was hopeful to learn how other developers and designers approach their appeals. Certainly there were many fewer people here than in the morning session. This was a bit sad, especially because the business model that TopCoder relies on to generate income is based on their ability to design, develop and deliver components to their clients. Anywho, when the round started, nicka81 (the TCO 05 design champion) was leading the designers, in first place over adic (the TCCC '05 champion) by 5 points (a large lead), with Pops in third, lagging by another 4 points. The development finalists were a lot closer, with colau leading sindu by less than a point, and visualage (the TCO 05 development champion) in third by also less than a point. Then the competitors opened up their reviews, the first time they've seen any of them. (They had already placed their bets before the appeals round even started.) As during the Algorithms semi-final, the designers' and developers' screens are shown to the spectators. Most of them read through an entire review, appealing their points one by one, and then moving onto the next review scorecard. By looking around, you could see many common design and development appeals:
Along the way, I saw some appeals that I myself wouldn't have made. One developer was clearly in the wrong (I saw him bring up his code) and yet appealed it anyway. Often as a developer, you feel each point deducted as a reflection of your personal integrity. It is very tempting to appeal each and every scorecard item that you didn't get a '4' on. But, you have to be truthful to the reviewers. In fact, today we saw quite a few appeals get rejected (and not surprisingly, some appeals succeed). On the other hand, I did see some really good appeals. aubergineanode appealed an item which said he didn't describe the thread safety in the interfaces he proposed. His response was right-on, because some components didn't have thread safety requirements, so if he were to specify it in the interface it would be too restrictive in the implementations of that interface. After 30 minutes, the top two developers hadn't made ANY appeals. In fact, after 60 minutes, sindu still hadn't made any, which was surprising. (He later made some, and his score increased accordingly.) I picked up on a conversation between ivern and antimatter about The Digital Run, which has an interesting "Rookie of the Year" concept. I myself was thinking about re-entering the design competitions, for which I might be a candidate for the designer ROTY (hmm, reminds me of ROTJ. Maybe I can pull off the Star Wars motif after all...) antimatter had the opposite point of view, that over the summer the development projects would be attractive to him (for the money), and then he might be eligible for the development ROTY. More interesting appeals that I saw included:
While the reviewers posted their responses, after almost 2 hours, the relative placement of all the competitors was the same. Lesson learned: appeals are a hard way to improve your score; the best way is to design or develop a solid submission in the first place. Latr dayz -- dplass P.S. I saw Pops' favorite bug appear multiple times: timeout in the appeals arena -- you submit your appeal but your session has timed out. Competitors quickly found a defense to this, by pre-writing their reviews in Word or Notepad, and then pasting into the form. This way, if they get the (error 500), they haven't lost their text. Maybe they'll fix it for next year! |
    |