pllk's blog

By pllk, 6 years ago, In English,

Here are my predictions for the IOI 2014 results. They were calculated using information from eduardische's IOI Database and Codeforces API.

Prediction 1 (current rating)

Prediction 2 (maximum rating)

Prediction 1 is based on contestants' current rating on Codeforces. If the rating is unknown, the default rating is 1500. The contestants are sorted according to the ratings, and then the IOI medal allocation algorithm is used. Prediction 2 is calculated in the similar way, but the ratings are maximum ratings.

There are some obvious weaknesses in the model (e.g. many ratings are missing, Codeforces is not IOI), but let's see how accurate the predictions are after the IOI...

Update: changes in teams from Colombia and Egypt

 
 
 
 
  • Vote: I like it
  • +15
  • Vote: I do not like it

»
6 years ago, # |
  Vote: I like it +13 Vote: I do not like it

the top 5 are same according to both predictions. let's see if this will come true or not. :)

  • »
    »
    6 years ago, # ^ |
      Vote: I like it +5 Vote: I do not like it

    I think that only 2 of this top5 will be in top5 on IOI.

    Thank you for interesting stats, pllk)

    It also will be great to have some live standings with CF nicknames (like this one for ACM ICPC).

    • »
      »
      »
      6 years ago, # ^ |
        Vote: I like it +5 Vote: I do not like it

      I think it'll be more in the top 5, but high places (top 50, let's say) will get messed up completely. Most importantly, there are many 1500s and some of them can easily take these spots with a bit of luck.

    • »
      »
      »
      6 years ago, # ^ |
      Rev. 2   Vote: I like it +17 Vote: I do not like it

      You nail it, man! scott_wu and KAN are in top 5. This implies that getting to top 10 on Codeforces pretty much guarantees a top place in IOI (pardon me, I'm being Captain Obvious here lol)

»
6 years ago, # |
  Vote: I like it +13 Vote: I do not like it

Haha, some candidate masters and masters are just unpredictable.

»
6 years ago, # |
  Vote: I like it +5 Vote: I do not like it

In my opinion, it would make more sense to call this something like rating statistics instead of results or predictions. The difference in format between IOI and Codeforces/Topcoder is quite considerable so that predicting IOI results based on solely the ratings is pretty much unreliable.

  • »
    »
    6 years ago, # ^ |
      Vote: I like it +5 Vote: I do not like it

    Well, these are predictions of IOI results, not just random statistics of Codeforces ratings. One purpose of this is to see how strong the correlation between IOI results and Codeforces ratings is.

»
6 years ago, # |
  Vote: I like it 0 Vote: I do not like it

CLGT!

»
6 years ago, # |
  Vote: I like it 0 Vote: I do not like it

You are too naive.

»
6 years ago, # |
  Vote: I like it +7 Vote: I do not like it

Hah, those predictions simply contradict my "rating medal cut-offs" which I stated there: http://codeforces.com/blog/entry/12950 :D. But this contradicts them in a good way, I mean, it says that achieving a good position in IOI is actually more valuable achievement :).

»
6 years ago, # |
Rev. 3   Vote: I like it 0 Vote: I do not like it

I really hope you get accurate with me. LOL

»
6 years ago, # |
  Vote: I like it +15 Vote: I do not like it

So glad this prediction wasn't correct for me (got 10th place). I guess high rating in CF guarantees a good place, however not-so-high rating doesn't guarantee a bad place :)

  • »
    »
    6 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Wow, congratulations on your outstanding performance!

    Could you describe your solution to holiday? Or is anywhere that solution described?

    I thought I solved it in O(n log^2 n) and it seems to give correct answers (using some funny divide and conquer), but because of some weird reason, it works ~10-12 sec for n=3000 :P.

    • »
      »
      »
      6 years ago, # ^ |
      Rev. 2   Vote: I like it +1 Vote: I do not like it

      I am not certainly sure whether my solution is correct, it runs in O(NlogN) but it may be a result of bad tests.

      Author solution may be in the website, since they posted the problems, I am not sure.

      • »
        »
        »
        »
        6 years ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        I also had an NlogN soln which passed in contest, but now I'm not sure if it actually works.

      • »
        »
        »
        »
        6 years ago, # ^ |
          Vote: I like it +1 Vote: I do not like it

        Excited enough to write about success on a cf (nothing wrong with it), but lazy enough that even combined with that excitement, don't want to post solution :D.

        Unexpectedly it turned out that problemsetter of that problem is my friend :P. Moreover, it turned out that my solution as exactly the same as his and it runs in O(n log^2 n). It doesn't seem that any official solution will be published and even if they will be, it will likely be the solution I already know :P. Since your solution is significantly faster than the author's one, I think it should be interesting to read about it.

        • »
          »
          »
          »
          »
          6 years ago, # ^ |
          Rev. 2   Vote: I like it 0 Vote: I do not like it

          It's not about laziness, it's just that I have testcases where my solution would fail, and I'm quite ashamed of that. Without the solution I still get the gold but its somehow sad to see bad tests on IOI.

          • »
            »
            »
            »
            »
            »
            6 years ago, # ^ |
              Vote: I like it 0 Vote: I do not like it

            Ah, OK. It's a significant difference between "I'm not sure whether my solution is correct, but it got 100 pts" and "I got testcases where my solution would fail" :P

            • »
              »
              »
              »
              »
              »
              »
              6 years ago, # ^ |
                Vote: I like it 0 Vote: I do not like it

              Yes, but some time passed between my comments and I managed to find bad testcases :/

»
6 years ago, # |
  Vote: I like it +51 Vote: I do not like it

I hope these charts will answered the question "How the CF-rating influence the IOI results?" :)  Large charts: CF(current) -> IOI2014, CF(max) -> IOI2014

  • »
    »
    6 years ago, # ^ |
      Vote: I like it +5 Vote: I do not like it

    Good work! And what are the correlation values?

    • »
      »
      »
      6 years ago, # ^ |
        Vote: I like it +3 Vote: I do not like it

      Correlation CFmax to IOI results = 0.731
      Correlation CFcurr to IOI results = 0.724
      Thus between the rating and the results there is a quite strong correlation (>0.7)