Silver_Fox's blog

Codeforces <-> AtCoder rating correlation plots and convertation — get your expected rank on the site you are new with!
By Silver_Fox, history, 3 years ago, In English

Hello, Codeforces!

I'm new on AtCoder and was interested, what rating would I normally get there after participating in CodeForces rounds, and how does rating on these two sites correlate.

Of course, there is a correlation ;)

Link to some plots and dependencies:

Rating correlation plots

preview:

And now you can get a prediction for the site you are new to.

Here is CodeForces <-> AtCoder rating converter:

Rating converter

preview:

You can send any suggestions to this e-mail:

[email protected]

  • Vote: I like it
  • +297
  • Vote: I do not like it

Comments Comments (16)
| Write comment?
»
3 years ago, # |
  Vote: I like it +6 Vote: I do not like it

Thanks to misha_sh for help with this!

→ Reply
»
3 years ago, # |
  Vote: I like it +15 Vote: I do not like it

I took AtCoder accounts which have link to corresponding CodeForces handles. Then selected only those who have at least 7 writed contests on both.

It may take much longer than 7 contests for AtCoder rating to converge. My AtCoder account https://atcoder.jp/users/ssvb/history currently participated in 10 ranked contests and has provisional rating 654. Average of the numbers from the "performance" column is 795 (~22% higher than the provisional rating). And I have no idea how many additional contests are necessary for my AtCoder rating to stop being considered provisional.

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

    Yes, thanks, it really changes the result a bit. Something like +60 to result when converting CF to AC rating, so the difference is not so big (updated in converter)

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

      According to this it will take at least 10 rounds for it to converge, for me personally it took 12 rounds for the warning of "rating not converged yet" to disappear.

      → Reply
  • »
    »
    3 years ago, # ^ |
      Vote: I like it +52 Vote: I do not like it

    Several aspects of the AtCoder rating system are briefly described in "rating.pdf" in the dropbox linked as "AtCoder's Rating System" on their front page. The most important thing to note is that the displayed user ratings are less than the internal weighted-performance-average used for calculation purposes, according to the following formulas:

    $$$ \text{displayed rating} = \text{performance average} - f(\text{number of rated contests}) $$$
    $$$ f(n) = \frac{F(n) - F(\infty)}{F(1) - F(\infty)} \times 1200 $$$
    $$$ F(n) = \frac{\sqrt{\sum_{i=1}^n 0.81^n}}{\sum_{i=1}^n 0.9^n} $$$

    Note that $$$ f(10) \approx 157 $$$ and even $$$ f(30) \approx 15 $$$, so this really ought to be accounted for explicitly unless you limit consideration to only accounts with quite a lot of rated rounds. Codeforces started using a similar system last year, but it has $$$f(n) = 0$$$ for all $$$n \geq 6$$$, so no such correction should be necessary.

    AtCoder also applies some monotonic transformation to ensure that all displayed ratings are positive, but I don't have any knowledge of what transformation that is, exactly.

    → Reply
»
3 years ago, # |
  Vote: I like it +14 Vote: I do not like it

Thanks great project ! Tho I am interested to know the formula / maths behind the conversion.

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

    I think the author used something like "Linear Regression" to find out the correlation.

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

      Something like that, since it's the easiest and the most working solution.

      → Reply
»
3 years ago, # |
  Vote: I like it +15 Vote: I do not like it

This mapped my max CF and AC rating pretty closely. Nicee

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

There is an interesting blob at 2100 Codeforces — 2000 AtCoder.

I hope I will get there someday.

→ Reply
»
3 years ago, # |
  Vote: I like it +18 Vote: I do not like it

I instantly recognize jh05013.

→ Reply
»
3 years ago, # |
  Vote: I like it +5 Vote: I do not like it

Got such a big difference:

ATC -> CF: 1537 -> 1928 (lol)

CF -> ATC: 1603 -> 1108

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

Pretty correct ig. But tbh it is easier to get rating from ABC/ARC and sole div.2 from the sites, making it harder to determine.

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

It seems that my AtC rating is 150 lower than expected lol

→ Reply
»
3 years ago, # |
  Vote: I like it -10 Vote: I do not like it

For me, the predictions were almost accurate.

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

I must point out that AtCoder did something to avoid negative ratings.

So AtCoder ratings in $$$[0,400]$$$ is actually $$$[-1200,400]$$$.

Anyway,it's a great project.

→ Reply