Arpa's blog

IOI vs CF performance: 2024 Edition
By Arpa, history, 4 months ago, In English
CF ratings are unrelated to your IOI results. CF contests differ significantly from IOI contests.

❌ Wrong!

A higher CF rating, with a gap of at least 100, correlates with an 87% chance of achieving a better placement at IOI. In fact, the relationship between IOI placement and CF rating is strongly connected. The Pierson Correlation Coefficient for these two factors is -0.83.


Correlation between CF rating and IOI placement

❗ What is that outlier participant on the top-right? He is Jiyu Shen (PubabaOnO), with CF rating of 3000+, but it seems that he was unable to attend the contest 💔

Here are the statistics on CF ratings for each medal:

Medal Minimum Median
🥇 2029 2469
🥈 1501 2274
🥉 1637 2082

Additionally, I've discovered an interesting correlation: participants' ratings one year before IOI also correlate with their final ranks. By the way, some participants showed remarkable jumps in their ratings during that one-year period. Specifically, there were 4 participants with 1000+ growth in CF rating!



Correlation between CF rating a year before IOI and IOI placement

Based on this insight, I’ve trained a model that can predict your placement at IOI. If you’re curious about your predicted placement for IOI 2025, leave a comment! Note that as time passes, the model can determine your place more accurately. So you may comment again 6 months later to see the updated prediction.

Check out the similar blog for IOI 2019 here.

  • Vote: I like it
  • +1106
  • Vote: I do not like it

»
4 months ago, # |
  Vote: I like it +39 Vote: I do not like it

What I find quite interesting is that difference between gold and bronze is only 400 points. Not a small amount, but it's by no means a large gap. I'm interested if something similar is at their OI level?

→ Reply
  • »
    »
    4 months ago, # ^ |
    ← Rev. 2   Vote: I like it +53 Vote: I do not like it

    I do believe being Red from CM is much harder than being CM from nothing. 7 out of 100 total CF user is CM, 4 out of 100 (+1900) CF user is Red.

    → Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it +36 Vote: I do not like it

    Reaching 2400 from 2000 is several times harder than reaching 2000 from 1600. For most people, it's nearly impossible. That's why the significance of a 400-point difference depends heavily on the rating range.

    → Reply
»
4 months ago, # |
  Vote: I like it +88 Vote: I do not like it

just a small detail, I believe Jiyu Shen was disqualified, but his performance was top 5 if I recall correctly, so it does still support your point

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it +10 Vote: I do not like it

    Thanks for pointing that out. I wasn't aware of it. I noticed his scoresheet was empty and assumed it was due to his absence.

    → Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it +19 Vote: I do not like it

    No he wasn't top 5.

    → Reply
»
4 months ago, # |
  Vote: I like it -102 Vote: I do not like it

"unable to participate" kekw, how much did you get paid

i guess he couldn't do it without a phone, sad to see the rewriting of the history going underway

→ Reply
»
4 months ago, # |
← Rev. 2   Vote: I like it 0 Vote: I do not like it

Please predict my placement. Lol, hoping for at least bronze :sob:

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    177!

    → Reply
    • »
      »
      »
      4 months ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      Wow, that is literally the last person who got bronze. Clutched haha

      → Reply
»
4 months ago, # |
  Vote: I like it +23 Vote: I do not like it

What if you remove everyone that hasn't done, say, 5 rated contest in the last 6 months? Otherwise, your data contains ratings that probably aren't "true".

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it +42 Vote: I do not like it

    I only included people who participated in at least a contest. I did what you say right now, and correlation got stronger just by 0.02 (=-0.84).

    → Reply
    • »
      »
      »
      4 months ago, # ^ |
        Vote: I like it +11 Vote: I do not like it

      I'm a bit confused about what is going on your plots with the rating distribution of the gold medalists. For example, it looks like only in this last plot is there someone with rating just below 2000 in gold. But this plot should simply contain a subset of the data from the other plots. So that doesn't make any sense.

      → Reply
      • »
        »
        »
        »
        4 months ago, # ^ |
          Vote: I like it +26 Vote: I do not like it

        Thanks for reporting. There was a bug in calculating this plot. After recalculation, it added just 0.01 to the correlation. Here is the correct one.

        → Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Whats mine im curious

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    115! But personally I think you can easily reach silver.

    → Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Nice pfp brother

    → Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

mine?

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it +2 Vote: I do not like it

    107, but your account is quite fresh. I think you can easily reach silver.

    → Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Predict my placement too please

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    95, on the cut of silver and bronze.

    → Reply
»
4 months ago, # |
  Vote: I like it +12 Vote: I do not like it

Can you predict mine placement?

→ Reply
»
4 months ago, # |
  Vote: I like it +65 Vote: I do not like it

While the data does indicate correlation between CF rating and IOI result, I disagree that it's wrong to say "CF contests differ significantly from IOI contests". It is a very different style of contest: 2-3 hours and you must solve as quick as possible vs. 5 hours, no penalty for resubmission, subtasks etc.

Of course, being good at one will probably mean you are good at the other, since the core problem-solving skills are applicable to both. But personally I struggle to consistently perform on div 1s due to silly mistakes, running out of time or getting stuck on one problem for too long and having too high penalty even if I solve later problems. Meanwhile in an OI contest none of those factors are nearly as significant, and I had a very strong performance at IOI this year. Of course maybe this just means I need to train more on Codeforces...

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

I am interested!

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    135

    → Reply
    • »
      »
      »
      4 months ago, # ^ |
        Vote: I like it +9 Vote: I do not like it

      I actually got 199 :(

      → Reply
      • »
        »
        »
        »
        4 months ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        Please note: The model doesn't predict the placement in IOI 2024, it also take into account the growth you'll have during the time you have until IOI 2025.

        → Reply
        • »
          »
          »
          »
          »
          4 months ago, # ^ |
            Vote: I like it 0 Vote: I do not like it

          Oh I didn't notice. That's nice. I also think that my rating actually says that I should rank better than I actually did in IOI

          → Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Predict mine please

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please ?

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

I don't even have time to compete in CF rounds though!!!!!

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Could you please predict mine?

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

would like to know mine over the course of time. grinding for expert rn

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Mine plzzz

→ Reply
»
4 months ago, # |
  Vote: I like it +9 Vote: I do not like it

please predict mine

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

mine

→ Reply
»
4 months ago, # |
  Vote: I like it -11 Vote: I do not like it

so Orange or higher in cf = almost guaranteed IOI medalist

quite interesting, cf rating is so much valuable than what I thought

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it +24 Vote: I do not like it

    cf rating is so much valuable than what I thought

    It's not. It's just that when reahing 2100+ level (preferably, from div1, reaching master from one lucky div2 contest isn't what i really mean...) you have enough skill/common sense, call it whatever you want, to be able to get medal even in you worst performance case. That's why it's (almost) always guaranteed IOI medal. You can't describe any medal with certain required cf rating (there is an actual CM gold medalist this year as well as there are many GMs in silver/bronze), but you can give such a "lower bound" for skill level (which div1 masters characterize pretty well). Everything else is mostly luck.

    → Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    the classic "correlation does not imply causation".

    → Reply
»
4 months ago, # |
  Vote: I like it +19 Vote: I do not like it

How does the fit look if you only look at silver and golds? I feel like getting bronze on IOI is fairly easy once you have some solid CP base, so it probably correlates with higher rating on any website. For the distinction of silver vs gold, though, it becomes a lot more critical to practice the format and the problem styles, so something like CF shouldn't be too predictive (except maybe in the 3000+ rating, which just signifies insane abilities overall).

→ Reply
  • »
    »
    4 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    a lot of it is also just random noise due to ioi being only a 2 day event. Even previous IOI results won't correlate with future IOI results. For example, consider https://stats.ioinformatics.org/people/1144

    → Reply
    • »
      »
      »
      4 months ago, # ^ |
        Vote: I like it +8 Vote: I do not like it

      I'm well aware that there's a lot of variation (me), but still might be somewhat interesting to see the correlation. But you're probably right that there's too much noise to get anything useful from this one year.

      → Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Could you predict mine, please?

→ Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

What about mine?

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    254

    → Reply
    • »
      »
      »
      3 months ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      What's the significance of this score?

      → Reply
      • »
        »
        »
        »
        3 months ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        It's your predicted placement in IOI 2025.

        → Reply
»
4 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please?

→ Reply
»
3 months ago, # |
  Vote: I like it -21 Vote: I do not like it

Hope you to calculate Guo Yuchong,Huang Luotian and Shen Jiyu's real score,not the score after they were punished to get a more precise answer.

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

What's your prediction about me?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Mine?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

mine ?

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    I don't think my model can predict correctly for unrated account. Go participate in at least 5 contests and come back.

    → Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

mine?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Mine please :)

→ Reply
»
3 months ago, # |
  Vote: I like it -30 Vote: I do not like it

CF contests are ICPC styling which is harder than IOI. Hence, a high CF rating means higher performance in IOI. Nice blog <3

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine

→ Reply
»
3 months ago, # |
  Vote: I like it +3 Vote: I do not like it

Can you predict my placement?

→ Reply
»
3 months ago, # |
  Vote: I like it +50 Vote: I do not like it

Can you predict mine?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Which is my prediction?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Which is my prediction?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

can you help me check my predictions?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

please predict mine

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please

→ Reply
»
3 months ago, # |
← Rev. 3   Vote: I like it 0 Vote: I do not like it

Curious, Arpa:

What does this analysis (I mean the predictions part) look for other IOIs? [Since you mentioned you can predict performance in IOI 2025, that would probably mean older accounts get an "unfair advantage" due to more time to practice?]

Also, this analysis has been done for CF ratings at the time of IOI, right? [Which may not be the max rating of the said account till IOI time]

I wonder how the graphs / plots look if you consider the max ratings?

Also, just for fun: can you predict mine please?

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Do you mean in IOI 2026+? I think that there is no accurate prediction.

    Here is the plot for max rating, but note that it shows a bit less correlation with the IOI result (Pierson Coefficient is 0.80). It seems very similar.

    The last contest you participated in is very old and can't be used for prediction.

    → Reply
    • »
      »
      »
      3 days ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      Actually, I meant something like IOI 2024, 2023, 2022 etc by "other IOIs"

      I basically meant that, since older accounts get "more time to practice" (example: someone's last contest is in 2023, so they get two years for IOI 2025 compared to someone whose last contest is in 2024), so perhaps would be better to think in terms of IOI of <"last contest" + 1 year> or something.

      Regarding the max rating plot, that's also interesting how it has very slightly lower correlation, but still strong.

      → Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you tell us how you predict placement of people ? Is that possible to make a website that can predict placement of users by their handle ?

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it +1 Vote: I do not like it
    Is that possible to make a website that can predict placement of users by their handle ?

    I'm thinking about that.

    → Reply
»
3 months ago, # |
  Vote: I like it +8 Vote: I do not like it

What about mine?

→ Reply
»
3 months ago, # |
  Vote: I like it +8 Vote: I do not like it

Can you please predict mine?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please?

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

I think the following adjustments would make it more accurate:

1) filter to those who have participated in at least six contests, because otherwise their rating would be downwardly biased due to the CF base rating of 1400 being credited over six rounds, and

2) filter to those who joined at least once over the past six months

→ Reply
»
3 months ago, # |
  Vote: I like it +1 Vote: I do not like it

me tooooo....please

→ Reply
»
3 months ago, # |
  Vote: I like it +1 Vote: I do not like it

Can you predict mine?

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    90

    → Reply
    • »
      »
      »
      3 months ago, # ^ |
        Vote: I like it +21 Vote: I do not like it

      Close, I am actually 89th.

      → Reply
      • »
        »
        »
        »
        3 months ago, # ^ |
          Vote: I like it +1 Vote: I do not like it

        Please note: The model doesn't predict the placement in IOI 2024, it also take into account the growth you'll have during the time you have until IOI 2025.

        → Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Could you calculate it after 2100? I am actively thinking about leaving codeforces

→ Reply
  • »
    »
    3 months ago, # ^ |
      Vote: I like it +8 Vote: I do not like it

    2100 is where the fun begins

    → Reply
    • »
      »
      »
      3 months ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      Lmao I agree. But sadly olympiads aint no fun and implementation and algorithm heavy

      → Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

It's funny that what used to be people asking for their future rating to be predicted turned into people asking for a prediction of their expected IOI ranking XD

Now on a more serious note, great job for your work, now as a follow up, it would be interesting to see whether this metric remained the same over the years and how can these results reflect a country's improvement or downfall in terms of future IOI performances and what can we learn from this.

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Whats mine im curious

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please.

→ Reply
»
3 months ago, # |
  Vote: I like it 0 Vote: I do not like it

Can you predict mine please

→ Reply
»
4 weeks ago, # |
  Vote: I like it 0 Vote: I do not like it

Predict mine please

→ Reply
»
7 days ago, # |
  Vote: I like it 0 Vote: I do not like it

I'm not an expert in these areas but it might simply be correlation and not causation. meaning that practicing cf contests might not directly improve your IOI performance (though it probably does a little bit, or up to a certain point), but people with higher ranking at IOI might have been more skilled at solving cf problems.

→ Reply
»
7 days ago, # |
  Vote: I like it -20 Vote: I do not like it

mine?

is it worth participating in IOI ??

→ Reply
»
7 days ago, # |
  Vote: I like it -13 Vote: I do not like it

We need 2025 edition now

→ Reply
»
7 days ago, # |
  Vote: I like it 0 Vote: I do not like it

Mine?

→ Reply
»
7 days ago, # |
  Vote: I like it +3 Vote: I do not like it

Btw Jiyu Shen participated in the contest. He would have gotten the gold medal if he hadn't been disqualified.

→ Reply
»
7 days ago, # |
  Vote: I like it +10 Vote: I do not like it

Mine?

→ Reply
»
6 days ago, # |
  Vote: I like it +4 Vote: I do not like it

What about mine plz qwq?

→ Reply
»
6 days ago, # |
  Vote: I like it +4 Vote: I do not like it

mine?

→ Reply
»
3 days ago, # |
  Vote: I like it -13 Vote: I do not like it

BTW Arpa,

I think there might be one more thing to take into account, especially at higher medal levels like Gold and all.

So, I would guess that most of the IOI people would be training on IOI-like problem sets (of course they probably often also train on CF, to maintain speed / form, different problems, just for fun and practice etc)

That said, often you'd find that some IOI people, especially at the top of IOI, like Gold etc, have a lot of skill, but it is not THAT much reflected in their CF rating, maybe because they might have done too few CF contests, so ratings didn't converge. Or maybe their "IOI skill" is high, but "CF skill" lags behind.

So, because of that, it might not be fully correct to say something like: "2469 CF rating roughly means IOI Gold level".

Of course I agree that "general CP skill" is a thing and CF is a good measure of that, so that is why the fact that "2100+ master = almost guaranteed IOI medalist (at least bronze)" is probably true.

I just wonder if we can do some sort of math, or something to account for this "lag" in "IOI skill" vs "CF skill" (or also the fact that some people's CF accounts lag behind due to less participations), and then get some sort of an "equivalent" of an IOI medal.

Example: maybe we can say: "while the median IOI Gold is 2469 rated on CF, something like +100 may be a better representation, so something like 2550 on CF would imply kinda skills comparable to getting Gold on IOI" Basically, IOI gold implies 2469 BUT we can't quite reverse this implication directly, more correct might be to say: 2550 CF implies median IOI Gold, or something like that?

I am just putting out thoughts here, and curious about your thoughts. Would love to see other ways of looking at this (or any other mathematical insights here)

→ Reply
  • »
    »
    3 days ago, # ^ |
      Vote: I like it -13 Vote: I do not like it

    Also, to be fair. cc: Arpa I just noticed there are data points in your plot which have like 1500-1600 or something rating and a bronze / silver medal at the IOI (As you mention in the rating cutoffs)

    I think this is nowhere near their true skill. It's just that they probably didn't do enough contests on CF.

    Wonder if some data cleaning here is needed.

    Also, your prediction model seems to map newbie / pupil level people also into some IOI rank like 250. (of course, some countries may be much much easier to qualify for the IOI from, and some like China are harder to qualify from than win IOI Gold), but still seems a bit too optimistic.

    → Reply
»
3 days ago, # |
  Vote: I like it -13 Vote: I do not like it

One more small potential thing to think about

cc: Arpa

This statement: "A higher CF rating, with a gap of at least 100, correlates with an 87% chance of achieving a better placement at IOI. "

Might be a bit of an overestimate, because given n people, it considers all n^2 / 2 pairs of ratings. Here, if some two points differ by say 500-600 rating or more, then they truly are in different leagues, so the stronger one probably WILL rank higher.

However, we shouldn't count such cases in the "87% success" cases perhaps, because it is obvious that someone rated 1900 will probably lose to someone rated 2500. This will trivially increase the 87% number.

The more important information is probably something like: "can a +100 or +150 jump in rating create a higher placement?". Probably considering all pairs with a rating gap of 100-150 points or something like that would be better?

Otherwise, a majority (or large fraction) of the N^2 pairs, would probably be with a > 500 gap or something, which skews things

Thoughts?

→ Reply
  • »
    »
    2 days ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    I can share all the data with you to investigate what you want.

    → Reply
»
2 days ago, # |
  Vote: I like it 0 Vote: I do not like it

mine?

→ Reply
»
44 hours ago, # |
  Vote: I like it 0 Vote: I do not like it

Mine?

→ Reply