TL;DR: As I mentioned in a recent comment, IMHO regular contests should be (at most) 2 hours long and keep longer contests for special/tournament events.
Personally, one of the main reason for me to compete in less contests in the recent years is that unfortunately I have less time to do programming contests. This becomes worse since Codeforces now have more and more contests longer than 2 hours. It's very hard for me to find more than 2 continuous hours for programming contest. I believe this situation might not be unique to me.
Below are the currently scheduled Div. 1 rounds.
| Name | Duration |
|---|---|
| Codeforces Round (Div. 1) | 03:00 |
| Codeforces Round (Div. 1) | 02:30 |
Below are the last 10 Div. 1 (or "combined") rounds. Only one of them was 2 hours long.
| Date | Name | Duration | # of problems |
|---|---|---|---|
| 09/18 | CodeTON Round 6 (Div. 1 + Div. 2, Rated, Prizes!) | 02:15 (was modified from 02:00 less than 24 hours before the contest) | 8 |
| 09/10 | Codeforces Round 896 (Div. 1) | 02:30 | 6 |
| 08/30 | Pinely Round 2 (Div. 1 + Div. 2) | 03:00 | 9 |
| 08/26 | Harbour.Space Scholarship Contest 2023-2024 (Div. 1 + Div. 2) | 03:00 | 9 |
| 07/29 | Codeforces Round 889 (Div. 1) | 02:30 | 6 |
| 07/23 | Codeforces Round 887 (Div. 1) | 02:30 | 6 |
| 07/11 | Codeforces Round 884 (Div. 1 + Div. 2) | 03:00 | 8 |
| 06/24 | CodeTON Round 5 (Div. 1 + Div. 2, Rated, Prizes!) | 03:00 | 9 |
| 06/18 | Codeforces Round 880 (Div. 1) | 02:00 | 6 |
| 05/28 | Codeforces Round 875 (Div. 1) | 02:30 | 6 |
Below are the distributions of Div. 1 (or "combined") contests since the start of Codeforces. The left-end is the start of Codeforces (2010), while the right-end is today. Blue line is "2h contest", while red line is "more than 2h contest". It is clear the trend is going for "more than 2h" contest. Better resolution image here.
Please bring back the old Codeforces with more 2-hour contests. I can't promise anything, but if future task authors would like to create a 2h contest, it will be much more likely that you will have one additional participant. :) If you are worried that 2 hours is too short for participants to solve your 6 problems, then save one problem for future rounds. You can then create more rounds. Win-win solution. :) Keep longer rounds for special/tournament events, like Meta Hacker Cup.
Call me boomer or whatever you want. I don't care.
PS: In case you are wondering, it took less than 2 hours for me to write this blog (including creating the chart).
Correction: The meme has a grammatical mistake. Should be "... have 5 problems and are 2 hours long".
Addition: I also want to argue that shorter contests might attract more "casual" participants. [citation needed]
If it's indeed talking about the past tense, then it should be :
But since these are still there, albeit once in a blue moon, it's technically correct to say,
I feel you bro.
Also 3 hour contests are very exhausting if they start in the evening or at night
Our lord has spoken
I fully agree with this opinion. Perhaps this will also help increase Div1 frequency too.
I'm generally opposed to this proposal: two hours doesn't feel like enough time to think deeply about problems, meaning that the only problems I can solve in such a round are those that I would have essentially gotten automatically in practice. This is especially relevant as an LGM, since I'm expected to solve an extra problem on top of what e.g. a GM/IGM would typically solve (and thus, it takes longer to get through the problems I can solve >99% of the time before I can work on the problems that actually determine my performance in the contest).
However, if this proposal is implemented, I think the way to do it would be to split the current Div. 1/Div. 2 rounds into three divisions. The problem is that while D1ABC are generally not hard enough to meaningfully distinguish between LGMs, they are time-consuming enough to absorb a significant fraction of the contest (e.g. in the most recent Div. 1 round, I was the fastest to solve ABC by a fairly wide margin and still needed over 40 minutes). This leaves limited time for the top contestants to think about the problems that are actually hard enough to distinguish between them.
Allowing top competitors (above somewhere in the 2400-2600 range) to skip D1A and B would (partially) solve this problem. No additional problems would need to be added at the hard end; I think four-problem rounds are fine given that very few (if any) people are capable of solving D1C-F in two hours anyway. In other words, Div. 2 would remain the same, the current Div. 1 would remain the same (or perhaps exclude the hardest problem), and the new division would include the current D1C-F.
I'm glad you posted this. An opposition gives me a new perspective. Thank you.
This is interesting. Instead of solving D1C-F for the top contestants, what about solving D1A-E (basically solving 5 problems like old CF rounds per my proposal)? Do you think that 1h20m is too short to distinguish top contestants with D1D-E?
I guess the question becomes: what to do with D1F problems. Well, occassionally the fifth problem can be D1F-level if the earlier problems are easier than usual. Or you can keep it for special events/rounds (like ICPC or IOI?).
Thanks for the response!
I actually think excluding D1F would make the problem worse. For all but a few competitors, solving D1A-E takes long enough that D1F is essentially irrelevant unless you skip an earlier problem to work on it, so removing D1F wouldn't significantly resolve the issue of not having enough time to think deeply about problems. However, it would remove a way for contestants to distinguish themselves aside from pure speed by excluding the strategy of e.g. solving D1ABC and then skipping to F, hoping to win over contestants who solve D1ABCD and don't have time for E.
sometimes number of problems more than 6 is overwhelming, 2 hours with 6 problems standard would be very nice.
I want to see one platform where it is an expected occurrence for some people (10%?) to solve every problem in regular contests. I think TopCoder tried to target this by trying to target 10% solves on the hard problem. (This is probably a controversial opinion, perhaps exacerbated by the narrative that a "perfect" ICPC contest has no team solving every problem. I don't particularly understand why it's bad for regular contests to have a nontrivial number of people solving the whole contest.)
Consequently, I feel like people extend the duration of the contest in an attempt to get more in-contest solves, but I speculate that this is a signal the contest is way too hard and should be adjusted somehow. (Is there a reason why regular contests don't strictly enforce a two hour timeframe?)
I think that contests where many contestants solve every problem are fine in general, but I do not think that it is suitable for Codeforces. The main reason I think this is because of penalty. I do not think that it makes sense for top LGM competitors to always be distinguished by implementation speed alone. I understand that with only a few problems per contest, there are many many ties that are broken by speed, but I think it is bad if, beyond a certain problem-solving level, speed becomes the only factor.
I think it is bad if, beyond a certain problem-solving level, speed becomes the only factor.
I think there's a reasonable argument to make that past a "certain problem-solving skill", speed is the primary thing being assessed. Note that differentiating by problem-solving skill is much easier at lower ratings. I also don't understand why people AKing a contest means that speed was the only factor to tiebreak, unless you're implying that problem-solving level and speed of solving problems are decoupled. Note that:
I think you have changed my mind to some extent.
My philosophy was that if someone is better than me, they should, on average, solve more problems than me rather than solve the same set of problems faster than me. The issue, however, is that in a contest with tens of thousands of people and only a few problems, many people will have the same set of solved problems no matter what.
I still feel like 10% is a very large amount of contestants to full-solve the problem set. I cannot imagine that the top contestants would find contests as interesting if they were essentially guaranteed an ak.
I think then that there are only two solutions that satisfy my very specific philosophy. The first would be to have OI-style problems with many many subtasks, but that is hard to implement in short contests. The second would be to follow Geothermal's suggestion and have many contests to appeal to all levels.
tl;dr: You are right, but I think that 10% is a very large number.
+1. I agree with the general premise of xiaowuc1's comment (that it isn't the end of the world if some people solve every problem), but I think 10% is an order of magnitude too high to target. Looking at the last few Div. 1 rounds, a top 10% rank leads to a performance in the mid-2700s, so this suggestion would essentially exclude problems over 2800 difficulty or so, which would make programming contests much less interesting for top competitors (and since everyone in the 3000+ rating range would consistently AK, for those contestants performance would entirely depend on speed, as GusterGoose27 suggests). It'd probably be very good for my performance, though :-)
(As an aside in case this hasn't occurred to you, what you're describing--a CF round cutting most problems above 2800 or so--is essentially a Div. 2 round. I enjoy participating in Div. 2 rounds out of competition, but I don't think Codeforces would be better off if we got rid of Div. 1 entirely.)
I agree that a problemset doesn't magically become better when you tack on a problem that gets zero solves, and I think a handful of people AKing a round (say between 3 and 8 AKs in a Div. 1 round) isn't a major problem. However, one advantage of including exceptionally hard problems is to allow for strategies involving skipping problems in order to differentiate yourself by solving an exceptionally hard problem (e.g. solving D1ABC, then attempting F in order to place ahead of people who could only work on D or E).
It might seem unrelated, but how did you make this chart? Don't downvote me to hell, please...
It is a normal question, especially considering the author bragging about doing everything in two hours. You shouldn't be downvoted, dude.
It's pretty simple actually. Since there are not too many contests, we can do some manual work and Google Sheet help instead of using APIs.
In case you are interested, here is the GSheet I used.
Not bad!
Here's an idea that's a little more out there but that might make jonathanirvings and I both happy. I'm not 100% convinced that this would work well as a contest format (and I'm very uncommitted to the proposition that all rounds should be like this), but I think it'd be interesting to try.
Take a typical combined round (or combine the problems of a D1/D2 round). Give contestants 7-10 problems to work on in two hours. Then score the contest based only on the hardest three problems each contestant solves. In other words, if you solve ACEGH, you only get points for E, G, and H. Use an AtCoder-style scoring system where earning more points always takes priority over time penalty.
Scoring would need to be handled carefully to reward strategies involving thinking about hard problems, even if this e.g. might allow time for only two problems instead of three. My general line of thinking is that lexicographical scoring (i.e., anyone who solves the hardest problem scores above anyone who doesn't, so e.g. H places above EFG) seems a bit harsh, but e.g. GH should probably place ahead of EFG (and perhaps FH should as well).
I also am not sure exactly how to handle time penalties, since it feels unfair to say that if you e.g. go for CDE but can't solve E and instead code B at the last minute, you should place at the bottom among contestants who solved BCD. Open to any ideas on how to handle this; maybe the implication is that contestants going for CDE should just type B early to hedge their risk in case they don't solve E, even if this would eventually slightly hurt their penalty among CDE-solvers.
This would shorten contests as jonathanirvings suggested while ensuring that strong contestants don't spend the bulk of their time on problems that are far too easy for them. It would also create interesting strategic considerations (for example, if I go in planning to solve EFG, but H turns out to relate to my strongest topic, should I change my plan and try to solve H and then work on whatever problems I have time for from there?) and would make working on the problems you find most interesting more viable as a competitive strategy.
I don't want to strategize on non-problem-solving part of the contest, and your format forces way too much of it to the contestants imo. Subtasks are already cancer enough in that aspect.
Judging by the upvote counts, the community seems to agree with you, so unfortunately it looks like this format won't be a solution.
no
Oh god, Please no. I dont want people (like me) to get free delta just because they were able to solve trivial problems fast and the others were unable to solve the harder ones in time. Take this recent 3hr contest for instance. https://codeforces.com/contest/1844
I finished A — D in first 25mins or so, and then bricked on the next problems for the remaining portion. My friendlist has around 20 people who solved E, and aside from the lgms who solved it instantly, around 15. Out of those, there are 7 solves after the 2 hour mark.
I would absolutely hate it if the fact that i can speedforce 4 trivial problems in 25mins ended up still giving me a big delta simply because the contest was too short (i still ended up getting exactly 0 delta). I firmly agree 2 hours simply isnt enough time to solve the "hard" problems for yourself.
At the end of the day, div 1 rounds are already infrequent, i dont think its a huge deal to find 3 hours of time maybe twice a month (and the rounds are also announced way ahead of time)? Especially coupled with the fact that most div1 participants are relatively more serious, so div4-div2 rounds might me more suited for "casual" participants.
Partially agree, as under most occasions I can solve all the problems I can solve in the first 1.5 hours or 2.
However, for some stronger participants it will cost more time for them to solve all the problems they can solve. The will and have to spend some time on the easier problems though.
As a result, I've got two other solutions to this problem that might work.
One is trying to make the problems with smaller gaps or more average gaps, which can cover more participants. Similar difficulty problems with different algorithms might also work.
Another is creating a Div 0 (or whatever you call) for stronger participants (e.g. 2600+ or 2400+) and change the problem number of each division to 5-5-4 or 5-5-5 (with one more problem with difficulty 1D). A small problem to this might be there are so few people rating 2600+ or 2400+ so rating calculation will be a hard thing.
Why, div1s are already rare enough so they are already kind of "special contests".
2 hours is too short to solve a real hard problem like 1E/1F
Thanks everyone for the interesting discussion!
I'm in favor of shorter contests personally, and I hear the opinion similar to this blog quite frequently. Thus, I'd say that the extreme redness of certain segments of the timeline, as shown in the post, is rather unintended. On the other hand, it's clear that a significant part of the community prefers longer contests.
In the future, for the regular rounds, we'll stick to the duration of 2 hours more diligently. However, certain rounds will definitely be longer.
before making such significant decisions you should at least hold an official poll (for div-1 users only) so that you have a more accurate measure of how many people prefer shorter rounds
+1. Given that posts both for and against shorter contests have received numerous upvotes in this thread, it seems ambiguous whether shorter contests are actually preferred by the community, and so it doesn't make sense to me to claim community support for this decision based on anecdotal evidence (especially since if CF currently favored shorter rounds, I'm guessing KAN would observe a similar number of requests for longer rounds).
This won't be very scientific, but let's try a poll now:
I'm in Div 1 and I prefer longer rounds
I'm in Div 1 and I prefer shorter rounds
I'm in Div 2 and I prefer longer rounds
I'm in Div 2 and I prefer shorter rounds
(To be clear, I don't think ">50% of people support this change" should be necessary or sufficient for CF policy to change, but the comment above indicates that community support is a major reason to have shorter rounds and I'm curious to see how much this change is actually supported.)
The reason I suggested an official poll which allows only div-1 users is because some people probably care about this enough to create a lot of fake accounts and skew the vote count of simple like button polls.
Sure, but I don't think the functionality to poll only Div. 1 users exists right now and I doubt it the administration will create it just for this, so I figured an informal poll would be better than nothing.
I love how you read a single blog and decide it all on your own.
I'd like to take this further, and ask why don't we have contests that last less than two hours. I'd certainly love to participate in 1.5-hour Codeforces contests for a change. The world, even Div1 world, does not have to revolve around problems for LGMs — and we are lucky to have AtCoder Grand Contests for that, anyway.
Topcoder coding format was "1 hour 15 minutes for 3 problems", and it worked well enough to distinguish the contestants.
As for the expected counterpoint of "results being too random". Yeah, a bit more random, but come on! If the contests always rank people perfectly according to their "global true skill", whatever that may be, what's a particular competition even about?
When Codeforces started, there were only div1 rounds with 3 problems in an ICPC-style contest. And it was amazing :)
As a coach, I personally prefer shorter contests with fewer problems. It becomes easier to read all the problems, discuss them with my team, and determine the best approach. When there's a large number of problems, it becomes challenging to find a dedicated period of time to discuss them all, and also some of the problems may go unread.
Now modern div1 contests have a lot of problems and last longer. For example, let's consider a contest that consists of 8 problems and lasts for 3 hours. What if we split these problems, ABCDEFGH, into two separate contests, each lasting 1.5 hours? The problems for the first half of the contest would be ACEG and the remaining problems, BDFH. What problems may the contestants have with this split?
personally, i feel contests have 3 types of problems for me.
1) problems i would most definitely get even on my bad days
2) problems i have a chance to get
3) problems i would never get
As you can see, my contest performance will be determined by 2 factors :
1) speed on problems of type 1
2) whether i could solve the problem(s) of type 2
In 90% of rounds, that type 2 group is extremely small, 1 problem 2 atmost, sometimes even 0.
Taking https://codeforces.com/contest/1870 as an example, i think ABCD are type 1, E is type 2, F is somewhere between type 2 and type 3, and GH are type 3.
If i got the BDFH set, my performance would be determined by how fast i solved the 2 trivial problems of B and D, which i dont like at all (even though i was very fast on both, and would benefit from such a system). [90 mins is too short for me to think and implement F imo]
If i got the ACEG set, and my performance would be the same as this contest, i would finish ACE in 73 minutes. This is far too close for my liking, especially when i comfortably solved ABCDE in the real contest.
So i guess, the main isssue im trying to explain is that you usually one "important" problem for you in the round, for example mine was E in codeton round 6. If you get the set without that problem, you are essentially competing for speed (due to the large difficulty gap). If you get the set with that problem, 90mins may be too small for you to solve that prefix.
I prefer large contests because they give me time to properly think on all "doable" problems.
I support making shorter rounds. I feel like all the arguments on both sides are just "I like it better that way" hidden by some attempt at objectivity. So instead I will explain why I like it better when the rounds are shorter: I don't like to sit for an hour (or more) without new ideas, and that's what usually happens in longer rounds. More time doesn't help to solve more problems in such situations.
I don't get what you mean by "such situation". Expected number of problems someone solves obviously go up as the round become longer.
The expected number of problems goes up, sure. I'm not so sure that going from "the expected number of problems solved by people in the top 50 is 5.01" to "the expected number of problems solved by people in the top 50 is 5.02" is a good reason to make me sit and pretend to think about the sixth problem for another hour.
For example, I don't think that this round benefited from another hour. Or this one. It's just my personal opinion.
First one you've linked had 32 out of top 50 participants solve an extra problem during the last hour.
Second one you've linked had 30 out of top 50 participants solve an extra problem during the last hour.
This looks significant enough to my eyes.
You got me, I don't really care about top 50.
I find this comment partially convincing. The underlying issue with the two rounds you provided is that a large number of top contestants (top 55 and top 26 in the two contests you linked) solved the same problems and were differentiated only by speed. This indicates that the sets were very imbalanced. In Pinely Round 2, for example, most top competitors solved the first seven problems in around 90 minutes, while none solved either of the two hardest problems in another 90 minutes. This shows that the difficulty gap between G and HI was far too large. The same is true of CodeTON Round 5: I is too hard relative to A-H (also, eight seems like far too many approachable problems to have in one round).
In my opinion, this is almost exactly as bad as a situation in which 55/26 contestants AK the round (in other words, I think the round would have been about the same quality if these unsolved problems were not used). The fact that nobody solved the next available problem indicates that it was hard enough that nobody/almost nobody had a realistic chance of solving it, meaning that the contestants who solved the same set of problems as rank 1 had nothing to productively think about for the rest of the round.
For these particular rounds, shortening the time limit to two hours would at least obscure this problem by reducing the number of top contestants solving the same set of problems. The problem is that shortening a round to two hours would create a similar problem if the round is better balanced so that top contestants are still thinking productively after the first two hours.
For example, in CodeTON Round 6, G and H were evidently approachable enough that strong contestants were probably thinking productively for the remainder of the contest. The same is true of Harbour.Space Scholarship Round: G and H were both within reach of strong contestants, and if the contest had been two hours long there would have not been enough time for more than a few contestants to think about them.
To me, the takeaway is that if authors are not confident that their round is balanced, the time limit should be short so that a large gap in difficulty between two problems doesn't leave a group of contestants (those able to solve the first N problems but with no chance at problem N+1) with nothing to do for the second half of the contest. However, if a round is balanced well, a strong majority of contestants should have plenty to think about and the round will be better with a longer time limit in order to reward those who solve the more challenging problems.
I don't agree to this line:
In CodeTon Round 5, I had a bug in my solution of I and I fixed it 5 minutes after the contest ends; which means I definitely had a meaningful progress in the last hour.
This is just an example, but I'd like to say that even if the standings look frozen, some contestants are trying to solve one more problem. No-AC doesn't mean no-thinking or no-enjoyment (at least for me).
if you know more time won't help you, you can close the pc and leave..
i checked your last 7 contests (i believe they are all > 2 hours, but if not even better).
In 3 out of your last 7 contests, you have accepted solutions after 2 hours. out of these 3, in one contest (https://codeforces.com/submissions/Um_nik/contest/1852) you have a nearly 2 hour gap between your last ac and the 2nd last ac. In the others, it is ~1hr (https://codeforces.com/submissions/Um_nik/contest/1864, https://codeforces.com/submissions/Um_nik/contest/1854)
The reality is that with balanced sets, this is quite likely to happen. Pinely round 2 and codeton round 5 are just examples of contests which are unbalanced at the highest level unfortunately due to a huge difficulty gap.
And what does that prove? None of the three contests you mentioned contradict my point. Moreover, among them only the Harbour.Space one I would consider a good contest in terms of my enjoyment. In it, I actually came up with something amazing and new (at least for me) during the contest (for problem H), which doesn't happen often.
The thing with trying to appeal to my personal experience to prove me wrong... I know my personal experience better. I also have public screencasts.
Screencast
At 30:28 I finished reading the three remaining problems and started thinking about D. It turned out that you needed to guess that reachable states in dp can be stored with O(1) information, which is the type of problem I hate. At 59:38 I abandoned D and switched to E, on which I made steady progress and started to write code at 1:31:34. I finished writing the first draft at around 1:58, and then it was ~20 minutes of debugging and rewriting logic in some tricky places.
The reason for long time before the last and 2nd last ac is sucky problem D (or me sucking at problems like D) and (less so) debugging time. I didn't make a breakthrough in any problem after 30 minutes of no progress on that problem. Moreover, I spent almost exactly 30 minutes on D before dropping it, and I needed almost exactly 30 minutes to come up with the solution for E (and I don't think there actually was even a 10-minute interval without any significant developments in the solution).
Screencast
At 51:00 I finished reading the three remaining problems and started thinking about G. It was a very nice problem with a lot of reductions to make, I'm actually surprised to see that I have done it in less than 18 minutes: I started coding at 1:08:24. There was a lot of code to write, and several tricky corner cases to consider, with all that considered I get AC at 1:39:06. I did not expect that I will be able to solve H, it reminded me of the AtCoder problem I couldn't solve, at 1:47:18 I went to read the editorial for that problem, it didn't help. Despite that, I came up with an amazing idea and at 2:07:05 I already started writing some code. I felt like I was missing something because the solution was too good to be true. Reaction to getting AC at 2:35:18.
Again, both those problems took less than 30 minutes to think about. Situations like H do not happen normally. I agree that this contest was super fun because it was longer. But I think that it is an outlier, not the norm.
Screencast
Problem E just sucks.
I finished ABC at 55:25. At 1:01:42 I started trying random stuff for E. At 1:26:56 I thought it didn't work, dropped E and switched to D. Started writing D at 2:03:13, and got AC at 2:15. Had only 15 minutes left, so finished the solution for E which I didn't think would work. Got AC 10 minutes later.
The reason for the long time between ACs is problem E.
I agree, it is hard to set aside such a long window of time. Lots of times, even 2 hours is a lot of time to set aside.
I think aside from the fact that I agree with the points in the blog, I will also add the fact that originally, CF rounds were two hours long and what was initially just an exception became a rule with the longer contests.
Maybe if there is enough demand for longer contests, we can have longer contests but with a different style as fundamentally speaking, two hour and three hour long contests are very different.
I guess something to solve both issues with the timing and the problem difficulty would be to reduce contests to 4 problems and 2 hours (problems 1-4 go to div3, problems 3-6 go to div2 and problems 5-8 go to div1 from a set of 8 problems).
Can anyone suggest, in what order topic should be selected by a beginner on codeforces. How to select topic. How to select question. How much time I have to spent on one question. It would be much better if you provide some references.
Quoted from this blog
Long duration is not always the solution.
If you're going to quote it, why not at least do virtual participation before doing so? It was a good experience for me to solve multiple medium-difficulty problems in 3 hours.
Good point. To admit, I have only seen the scoreboard and some of the problems briefly. I will try to find some time to do the contest virtually (which won't be easy for me due to the duration).
It didn't work to balance contest, but it does not mean that contest would have been better within 2 hours. Imo, it would be even more unbalanced in this case, and 3h duration was a correct choice, if we assume problemset was fixed.
I want more 2-hour contests because a contest with 3 hours means if I want to finish it, I won't go to bed until 01:35a.m. (because I am Chinese in UTC+8)
I agree with you. My timezone is UTC+7, so my parent always tell me to go to bed early. That's why I can't paticipate in many contest
SO TRUE !