if no then the "Everyone is welcome to participate." is misleading

Yes, you can participate in a special corporate scoreboard. Added information about that to post

First you need to register on ICPC Global. Then follow the link and find your account by nickname or email

No.

Maybe you have registered the same team before for ICPC. I faced the same issue and changing the team name worked for me.

You need to tick the box "I will be a coach and a participant" and do not enter yourself in the field of a team member

ICPCNews answer this question and above please

That's expected behavior. All teams will be accepted in closest time

There are now rated contests :)

Dear Nafis, the prize will be transferred to the bank account specified by the winner.

Yes, same email address — the screenshot is from a previous Huawei contest

Yes, they mentioned that we have to register to be able to get a prize.

Well, I guess because of following.

Assume $$$Q^{min}_t = 4$$$. Assume I do following: let the $$$[1, 4]$$$ be the segment of tenant $$$t$$$, and for query $$$t \; x$$$ I answer $$$(x - 1) \bmod 4 + 1$$$. Then, if input queries are $$$[t \; 1, t \; 5]$$$, then I print $$$1 \; 1$$$, and its considered as wrong answer, because I could find another free frame for page $$$5$$$.

Edited

Second this. Otherwise significant part of contestants' effort will be spent on black box experiments which is neither fun or educational...

Hi, thank you for paying attention to the Challenge. The tests, as well as details about their generation, are secret. The intention is to incentivize contestants to solve the general problem, and not a particular test set.

I second this. I kiiinda get organizers motivation and I don't want to say it's straight up bad decision, but not being able to reliably conduct any meaningful local tests was what swayed me away from engaging in this one

Now that the contest is over: testcases felt a bit like a guessing competition to me. Case 26 and 27 in particular felt very arbitrary. It turned out that you have to evict the most frequently used page here to score well. I don't see how hiding those cases could lead to a more desirable outcome for the organizer in terms of solution quality. You just submit random stuff and hope for a lucky hit that increases your score.

It depends not only on your number of page faults, but also on number of faults in the base solution (LRU). So if your SLA matches base answer, which is most likely also 10, you’ll get Cost=0 and Score=500

whiteapple is so strong!

Can you please do the same distribution for the last challenge?

Hello, in order to answer your questions in time, please log in to the competition question interface https://codeforces.me/contest/1813 and click 'ask a question'

About prize money, that's not a problem. As long as you have a collection account and the account can be paid through the banking system.

You can choose ICPC U as your institution in the ICPC System

15ms is a user (processor) time, but not a real time. Sleep increases a real execution time and doesn't affect the user time.

Try to replace 200 with a huge value: you should recieve something like "Idleness limit exceeded" (don't know the exact name of this outcome on CF).

Some of these are super depressing.

Yes, please use "ICPC U" as your institution. There is no way to link the CF account to ICPC account, so all matching will be by email — please use same email in both systems

yes, same email address — as explained in the comment directly above yours ;)

Yes, but you need to do that before 5th May — that's deadline for registration.

By the way, Codeforces/ICPC should have some strategy to prevent a handle from participating from multiple accounts (assigned to each of his family), just to get more prizes... And plagiarism detection is not enough...

BTW, one of them is known for getting top places in the previous Huawei challenges. But the second CF user was created just before this contest and this indeed looks a bit suspicious. Especially submissions' timing.

https://codeforces.me/submissions/whiteapple

ICPCNews, HuaweiChallenge, please answer. I'm really worried about it. I wrote messages to you to two emails, but did not receive a reply.

Yes, everything correct, your team was accepted — you can check that in list of teams

I hope after system testing they will enable the possibility to read others' solutions

I couldn't get past ~12k by dynamically swapping between LRU and LFU :( I guess I should've tried this more.

Btw, MRU did really well on tests 26 and 27 for me. So maybe you could try dynamically swapping between all three of them (my final submission was exactly this, but ig my swapping strategy was really poor).

PS: Did you mean "second chance" LRU? Plain LRU didn't seem to be doing well enough for 13.7k.

There are 3 algorithms that are actually useful : LRU LFU MFU (works for tests 26,27)

Taking the best one for each test case is enough, but you need a good strategy to decide between tenant. To decide between two different tenants my formula is : ((faults — faultsInSolution) / faultsInSolution)^2 * sign(thing on the left) * priority / (cache Used / Qbase)^2 This alone gives 14k (even 14400 I think) To gain the last points, you had to notice that in some tests, the structure was in blocks where two blocks of the same person are identical. I wrote a program that could see this case, and I ran belady's algorithm by looking at what happened in the past. This gives 500 on test 4 for example.

Let's define a few caching strategies: LFU, LRU, segmented LRU+LRU (those were mentioned in the statement), MFU (evict the most frequent one, testcase 26+27), some hybrid (page hits + 3e-3 * time of last hit), least frequent among the last few requests (I used 3 * tenant's current Q, but the factor doesn't matter that much).

You run each of these strategies on each tenant and keep track of what gets the most hits. This strategy then becomes your main strategy for this one tenant.

The decision which tenant to evict was rather painful, as whatever test I generated offline showed different behavior than submits on deciding whether or not a change was good. Here are some charts I made to see how close my prediction gets to the final SLA and such:

After some trial and error I settled on evicting whichever tenant minimized the Score() function below

private double Score(int qCurrent)
{
    double progress = (double)Solution.Time / Solution.QueryCount;
    double slaPlan = PredictSLAreal(qCurrent);
    double slaExpected = PredictSLAbase();
    if (slaPlan >= slaExpected) return Math.Pow((slaPlan - slaExpected) / slaExpected, 2) * Priority / Math.Pow(QBase, 0.9);
    return -1e-9 * Math.Sqrt((slaExpected - slaPlan) / slaExpected);
}

private double PredictSLAbase()
{
    double remainingFactor = ((double)Solution.QueryCount - Solution.Time) / Solution.Time;
    remainingFactor *= 0.1 * Math.Min(1, 5.0 * Solution.Time / Solution.QueryCount);
    double qAvg = 1.2 * qAverageLRU - 0.2 * QBase;
    return LRUStrategy.SLA + LRUStrategy.SLA * remainingFactor * (DatabaseSize - QBase) / (DatabaseSize - qAvg);
}

private double PredictSLAreal(double qCurrent)
{
    //if (Solution.Time > 60000 && ID == 1) Debugger.Break();
    double remainingFactor = ((double)Solution.QueryCount - Solution.Time) / Solution.Time;
    remainingFactor *= 0.1 * Math.Min(1, 5.0 * Solution.Time / Solution.QueryCount);
    double qAvg = 1.2 * qAverage - 0.2 * qCurrent;
    return SLAcurrent + SLAcurrent * remainingFactor * (DatabaseSize - qCurrent) / (DatabaseSize - qAvg);
}

Dynamic swapping between LRU LFU MRU and MFU )

Harmony_'s solution is still running...(On test 29 now)

It has been ~1h for it to be runnning.

Hope that won't take another 1h in system test.

By switching do you mean that you periodically look at the recent history and determine which algorithm would have minimized the miss rate (or another function) and you end up using just that one for the near future?

Dont know

Hi! We updated blog post: you still can register on ICPC system. Please do that till May 5th

Rejudge finished