Repeal the division cutoff change
Firstly, it causes rating inflation. In all div1 + div2 rounds the div2 players boost the div1 players, so even if someone does really badly relative to div1, he or she will get a much reduced loss, or even a positive rating change! In these div2 rounds there will be no reds/yellows to counter the buffer as well, so candidate masters will benefit even more!
Secondly, it does not solve the problem it seeks to address, rather it shifts it to higher rating ranges. Now instead of barely 1900s being unable to solve anything, we will have barely 2100 “masters” being unable to solve anything! We all know these titles are just for show, but this is a little ridiculous.
In addition, this shifting of the rating range also exaggerates existing problems. In rare occasions, even now, somebody will score 1st place in div2 and jump to master or high candidate master even if their skill level is low. With this change, we will have people jumping to grandmaster without even doing a single div1 contest!
With this change, ratings will become a joke. And that is why we need to repeal the division cutoff change.
Amendments to the creation of division 3
Thanks to ppavic's insight, I revised my statement on repealing the creation of div3. There still need to be some key changes, however:
The rounds should be limited to players with less than 1400 rating. 1600 is much too high if we are aiming to see the difference between beginner players who can solve only easy problems like A and B. If someone can solve A, B, C, then he/she should be in div2, as that is already solving 3 out of the 5 problems. In addition, we need a quota on the number of div3 contests in order to stop them from hampering the div2 or div1 contests, perhaps one per month.
Well, I see your arguments and I agree to some extent.
But in favour of Div 3. , for users which have no chance of solving problem C, Div 2. contests turn into a typing race and often a reading race, how fast can you read the problem statement and how fast you code it. In my opinion, Codeforces shouldn't be focused on that aspects but rather on logical thinking so I think having more diverse but still easy problems will help differentiate users with rating below 1600. And another thing, making Div 3. Rounds with easier problems will enable authors to put harder problems on Div 2. So I think we can expect a usual Div 2. B problem become an A problem so users with ratings > 1600 which mostly solve problem A without any problems won't know have to trouble themselves with problem A.
I looked at the most recent div2 round and found that those users in about the 1000th places take about an hour or more to solve only A and B. That means there's still 4000 players who take longer to solve A and B. It's only a typing race if you are high leveled, as in about 1500+ rating, but not good enough to solve C, as in < 1600 rating. So the "type race" will be an even bigger problem in div3!
In addition, this shows us that making easier problems isn't the solution, as 4/5 of the participants struggle to solve the two easiest problems. So div3 will be a very small benefit to those new users, if there is any benefit in the first place.
If 4/5 of the participants struggle to solve the two easiest problems, of course it's best to give them more easy problems
Then they will struggle to solve 2 easy problems out of 5 div3 problems.
Well not if you give them easy enough problems. The smaller the group of people which the contest is aimed at is you can set better problems for them because you can more easily tweak the difficulty of problems. In a regular Div 2. Round it is very hard to write problems which are challenging enough for users with > 1700 and especially > 1800 and are at the same time easy enough and interesting for users < 1600 especially < 1400.
In my opinion, the last round was an exception. But most of the time for beginners the strategy is same. Solve A, solve B and think about C mostly without any success. I agree there are some exceptions where A and B are of similar difficulty but I think we can agree that most of the time A is easier than B. But in a Div 3. round there is much more diversity. Now you are faced with probably 4 problems which are solvable for you and you won't lose interest after half an hour. And some users have a different skill set. Someone is really good at some kind of problems. Now with 4 problems, you will be able to show where you shine.
Sorry, I was looking at the VK cup round, not the most recent div2 round. I looked at a couple more rounds and I agree with this assessment now.
In this case, though, I still believe there should be some key changes. I think it the rounds should be limited to less than 1400. 1600 is much too high if we are aiming to see the difference between players who can solve only A and B. In addition, we need a quota on the number of div3 contests in order to stop them from hampering the div2 or div1 contests. I propose a maximum of one div3 contest every month.
As someone who a few months ago did the process "solve A, solve B, then think about C without any success", I think that div3 rounds will not be beneficial to beginners.
Being exposed to problems out of your reach forces you to improve. Shoving all these people into div3 will only further decrease time these people are able to spend on challenging problems.
yes but it can be demotivational as well, I know a lot of people interested in learning cp, but is SO common that they simply fail all problems on their first contests. I think this is not attractive for some people that are still getting to know cp, this passes the idea that you have to be some kind of genius to get into it, which is not true.
All this I am saying is experience inside my university.
"As someone who a few months ago did the process "solve A, solve B, then think about C without any success", I think that div3 rounds will not be beneficial to beginners."
you sound somewhat like my dad.
"Back in my day we had to grab a real man's book to learn something. This generation is lost with this internet stuff. Nowadays people are too lazy... Oh look, uncle beth sent me another life on candy crush!"
After all, if one enjoys solving problems, then they will participate, without caring if the contest is rated for them or not.
Personally, I'm much more interested about the quality of the problems and the ideas behind them, rather than ratings, divisions, etc.
Ratings are very important. The whole point of rating is to provide an accurate measure of players' relative skill levels. That's why codeforces has high quality problems, because problemsetting is restricted to skilled players. If people only cared about problem quality, then quality would actually decrease -- there would be no way to measure who is a good algorithmist, and everyone would end up solving a bunch of problems set by who-knows-who on obscure online judges.
I agree with you on this one. The quality of Codeforces is that it enables people to compare between themselves and in my opinion, it gives people a huge boost of motivation. When you have friends who are about as skilled as you are it is fun to compete with each other and sometimes even tease each other a bit :D.
Sorry, your comment makes absolutely no sense to me. I have some doubts regarding both
You don't need a contest to be rated to figure out who performed well there, and you don't need ratings to get feedback and metrics on the contest and its author (from other participants after the round, for example).
Moreover, if we're trying to talk about things seriously — having anonymous stuff (not knowing author's rating, ideally not knowing him at all) is better for more objective feedback. I guess if I prepare some trashy round, and some div2 guy prepare some trashy round, I'm going to get more upvotes and positive comments there just because of having red rating — which has nothing to do with my problem setting skills. Moreover, community is probably going to pick me or some other red coder over slightly better setter with bad rating, which doesn't sound good. Yep, simply because of people having the bias which you described.
If we don't have rating as an indicator of good problemsetting, then what do we have? Only solver feedback. We don't know if the solver feedback is legitimate, or if the problem is suited for high levels, unless we have ratings of the feedback authors. Maybe a bunch of lower-leveled players said it was too hard, but you don't know that because you don't know if they are actually low-leveled or high-leveled. Then mostly the feedback will become either invalid, or it will become too tedious to sort through manually. So you still need ratings to judge the feedback.
And in addition, without ratings, good problemsetters won't be able to demonstrate their skills. Old setters, who already have fame and feedback, will just keep churning out problems that most people solve, and new setters will have no feedback and no solvers, because they have no reputation. With ratings, someone can say: oh, this fellow knows how to do a contest, and even though he hasn't set any problems before, these problems are worth trying. How would you determine who gets to set a contest?
"In all div1 + div2 rounds the div2 players boost the div1 players, so even if someone does really badly relative to div1, he or she will get a much reduced loss, or even a positive rating change!" Well, when I did awful in a Div1+Div2 round, I got -91 rating. When I did awful in a Div1 round, I got -81 rating. I don't know where did you get the idea that div1+div2 rounds are good for div1.
I wrote "really badly relative to div1", not "really badly relative to div2". As in my case here.
I wouldn't say that was "really badly". There were more than 100 div1 users below you and about 350 above you, which would in a div1 round probably be about -20 rating, which isn't much reduced loss.
That would probably be more like -60 rating. Also, in your example, notice that you would have to solve A B and C in your combined round for it to be roughly equivalent to your div1 round.
Also, I would really appreciate it if you could tell me how you found the status of div1 only, thanks!
I just scrolled through the leaderboard and counted.
Your concerns are completely valid, but the statement "With this change, ratings will become a joke" is a little bit exagereted, people still need to be ahead of others to gain rating.
Also, if it is too easy for purples to jump to div. 1 doing div. 2+3 only contests I am sure MikeMirzayanov and his team will adapt it accordingly. A simulation of the rating evolution with the upcoming change would be cool.
Sure, they will need to be ahead of others to gain rating, but it is still the case that anyone who is between 1900 and 2100 rating gets a couple of hundred free points and everyone else doesn't benefit. It devalues the ratings, and it doesn't fix the main issue.
Even if all candidate masters reach 2100+ rating, with a couple of div. 1 rounds those (as you suggest) “unfair” points will show their true face and redistribute those “fake” rating points to the “true” div. 1 users.
But I bet this scenario will not happen. I believe you underestimate the difficulty to perform well in Div. 2, mainly when competing against other candidate masters.
No, they won't redistribute, because master will be the new candidate master. In every single div2 round there are tons of people who just make new 1500 accounts because they want to start at a higher rating again, and then they get farmed off of by the experts because they can only solve like 2 problems. With these changes the 1900-2100 people that participate in div2 will become 2100+ because of these inflations in the div2 round.
And this of course will happen, look how easy it is for combined div1 + div2 already, it will become even easier than that!
You can read about the rating inflation fighting point in Codeforces rating system. The reason that it didn't work well in combined rounds is that it only affected top rating contestants who got the very top ranks, and middle ranks (past
) got the boost. But with the first ranks going for candidate masters who would be a small portion of the participants, the adjustment would decrease the effect of the boost and prevent them from going too much off the boundary, and middle ranks would not get high (or maybe even positive) rating change in a Div2 only contest anyway.
Hmm, but with the maximum effect of this only being 10 points, I'm not sure if this makes enough of an impact.
Predicted