rng_58's blog

Rated range of AGC
By rng_58, history, 5 years ago, In English

AGC is our de facto Div1 contest, traditionally rated for everyone. However, given the difficulty of the problems, we think we should set a lower bound for the rated range of AGCs. As an extreme example, it was enough to submit anything (even a compile error) to get a brown performance in the last AGC.

Currently, there are three implicit "divisions" in AtCoder:

  • Div 1+ (2800 <= rating). Equivalent to CF mid-high reds.
  • Div 1- (2000 <= rating < 2800). Equivalent to CF low reds, oranges, and purples.
  • Div 2 (rating < 2000). Equivalent to CF blues and below.

And there are three types of rated contests:

  • AGC: Targeted for Div 1 people, especially Div 1+.
  • ARC (or orange-circled, sponsored contest): Targeted for Div 1 people, especially Div 1-, with two easy problems attached.
  • ABC: An educational contest for Div 2 people.

We are planning to make AGC rated for its target range, and the next AGC will be rated for >= 2000. Still, everyone can participate in the contest and they appear in the standings as usual; it's just a matter of rating computation.

P.S. It's possible to reach Div 1 in one ARC contest, but ARCs are not very frequent and in case some strong newcomer misses it, it takes 4 ABCs to reach Div 1 (and it feels a bit too many). There are various ideas, which do you prefer?

  • Keep it. Ask them to participate in 1 ARC or 4 ABCs.
  • Make AGCs rated for >= 1200.
  • Make AGCs rated if your internal rating is >= 2000.
  • Make AGCs rated if you get performance >= 2000 at least once.

UPD: For the next contest we'll go with the second option.

  • Vote: I like it
  • +208
  • Vote: I do not like it

Comments Comments (24)
| Write comment?
»
5 years ago, # |
← Rev. 2   Vote: I like it +45 Vote: I do not like it

Are you sure the rating division is correct? AtCoder rating 2000 seems to be definitely >= orange on CF (given the rating inflation on CF).

Personally, I can mostly solve at max 2 AGC problems in-contest, but it is fun none-the-less and would really prefer AGC being rated for me. I am 1652 on AtCoder currently after 11 contests. Performance >= 2000 once seems like an okay metric, but display rating >= 2000 seems too high.

→ Reply
»
5 years ago, # |
  Vote: I like it +70 Vote: I do not like it

I find that ABCs are too easy for me, while AGCs are extremely hard and I could hardly solve any problems in it. It would be better to hold more ARCs for participants like me, I suppose.

(In fact there haven't been any ARCs for at least two months, according to the 'Recent Contests' list.)

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

    We are planning 12 ARCs per year. The next one is tomorrow, and we'll have one more in two weeks. (sometimes we have a lot of problems, but sometimes we don't have problems, so they are not necessarily equally spaced.)

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

It may seem like ratism, but what about manually setting rating to 2000 (thus allowing participation in AGC) if red person writes you PM on cf?

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

    Bruh vote for last option. Ask that red to participate in ABC. Top 200 in ABC is enough for at least one >=2000 performance.

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

How about people who do not meet the lower bound criteria for a contest can opt-in before a contest to make it rated for them ?

→ Reply
»
5 years ago, # |
← Rev. 4   Vote: I like it +41 Vote: I do not like it

Personally I will prefer something along lines of >=1800 as lower bound for AGC. ARCs are as rare as AGCs. 2000 is just an upper bound of ABC. It will be more or less like. A slight drop in AGC and you are forced to participate in ABC in order to be eligible for AGC.
Another reason being I'm stable orange on CF but just above 2000 at atcoder. Div1- at atcoder according to you.
AGC is targetted for div1+ but please allow Div1- as well.

Edit -
I missed that you said you will allow out of the contest submissions. If that's the case I'm fine with any lower bound. Even 2400 will work for me. :P

→ Reply
»
5 years ago, # |
  Vote: I like it +46 Vote: I do not like it

Why ARCs have to be sponsored?

→ Reply
»
5 years ago, # |
  Vote: I like it +53 Vote: I do not like it

Make more ARCs please.

→ Reply
»
5 years ago, # |
  Vote: I like it -80 Vote: I do not like it

Make AGCs rated for >= 1200. is good because I am a bit over 1200.

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

I vote for 4th option. Because one bad performance in AGC shouldn't force someone to participate in ABC (as ARCs are quite rare).

→ Reply
»
5 years ago, # |
  Vote: I like it +32 Vote: I do not like it

I think judging people by performance is better. As you said that we don't want super strong newcomers miss it,it could be a good idea to make it rated for the participants whose highest performance>=2000

→ Reply
»
5 years ago, # |
← Rev. 4   Vote: I like it +15 Vote: I do not like it

I think there's 5th option! It's based on 2nd option but different, is like doing all of the following:

  • Make AGC rated for users of rating at least 1000 (or 1200 as another option).
  • Every time you hold AGC, hold AGC-based ABC (rated for below 2000) simultaneously.
  • (Optional) Either of 4th, 5th, or 6th problem of the simultaneous ABC will be same as AGC-A.

Let's define rating 0-999 range "Div 2-", and 1000-1999 range "Div 2+". The following will happen:
  • Div 2- coders are to participate in ABC in general, because they're not rated for AGC.
  • Div 2+ coders can choose participating in either of AGC or ABC, but when the AGC contest is difficult, or if not good at it, they will select ABC, and otherwise they will select AGC.
  • Div 1 coders are to participate in AGC in general, because they're not rated for ABC.

If we do like so, we can solve following problems!
  • Even if they solve 0 problems in AGC, their performance will reasonably decrease, if the easiest problem is easy enough.
  • We can mitigate "the cliff of rating 2000", which is caused by the fact that "what type of skills are required to increase the rating" is so different for 1999- and 2000+.
  • Especially in rating range 1000-1999, some prefer AGC, and others prefer ABC. But it will not confine the way of participation or rating evaluation, and both sides will be happy.
  • If a strong newcomer participate in just 1 ABC and achieve performance 2200+, he/she will be rated for AGC instantly.
→ Reply
»
5 years ago, # |
  Vote: I like it +2 Vote: I do not like it

I agree with the 4th idea, but "performance" may be not famous to newcomers. It would be necessary to appeal the system to them.

→ Reply
»
5 years ago, # |
  Vote: I like it +21 Vote: I do not like it

I would prefer the last option, let me explain a bit why: I have participated in 7 contests with >= 2200 performance in the last 3 (even one 2400 performance), and all of my deltas have consistently been over +100, however I'm still "only" at 1770 rating. I'm not here to brag, but I think I should correspond to the lower range of people that should be rated in the AGC, I can solve 2+ problems, and the problems are interesting and solvable, not impossibly hard. What I'm trying to explain is that if your level is around 2200 on Atcoder like I think mine is according to my "performance" column, you might need a not so small amount of contests to reach 2000, and since the contests are not very frequent, a lot of purple/orange people on codeforces might not have had the time to reach that rating yet.

These were just my thoughts, maybe I misunderstand some things, or I am completely wrong, but at least you have my perspective.

PS: I might not understand something, but I am very curious as to what the internal rating is, is there a way to see mine?

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

    How to calculate :

    While rating>400, (displayed rating) = (internal rating)-f(n) // n = times participated

    f(n)=1200.0*(sqrt(1.0-pow(0.81,n))/(1.0-pow(0.9,n))-1.0)/(sqrt(19.0)-1.0)

    If n=6, f(n) = 288.6... then your internal rating is (about) 2059.

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

I'm orange on Codeforces and I've done 5 ABCs with 1900+ performance each time and upto 2400 once. My rating is still less than 2000. I think the 4th option is probably the best.

And yes, I do find AGCs insanely hard.

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

Is this really an issue though? The last round just had an exceptionally hard problem A, compared to other recent AGCs. And anyway, anything below a green rating is virtually pointless, so what exactly is the issue if they got a brown performance?

→ Reply
»
5 years ago, # |
  Vote: I like it +63 Vote: I do not like it

In my opinion, the minimum rating you have chosen to make AGCs rated (that is 2000) is too high. Yes, it is possible to reach 2000 in one ARC contest if you are very strong, but if you are one of the many other reds this is not true.

My favourite option is Option 2. It solves the issue "it was enough to submit anything (even a compile error) to get a brown performance" (because if you are above 1200 then a brown performance is not good) and at the same time it is very easy for anybody in Div 1- to reach the threshold.

Regarding options 3, 4; I don't like them because they make it "complicated" to understand whether the contest is rated or not for a participant.

p.s. Among the contestants in the Top 10 of the atcoder ranking, 3 of them needed at least two contests to reach 2000. Among the contestants in the ranking range $$$[101, 110]$$$, only one reached 2000 in one contest (but it was a grand contest!).

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

Auto comment: topic has been updated by rng_58 (previous revision, new revision, compare).

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

I'm the only one who hates the first two problem of the ABC's?

There is nothing educational in solve two "a+b" problems.

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

Maybe making AGC rated for only >= 2000 rated coder is not good. Because someone who is rated less than 2000 can also perform a lot better than 2000 and then he would not get the result. Similar things happen in codeforces. Everytime we find some cyan coders performing in the level of CMs.

→ Reply