% of accepted is not unique over authors solved

[Destination Goa]

I don't know if this topic occured earlier.

Recently I've solved one 17% problem and then decided to beat the ranks on it. This means about 10-20 AC solutions with different timings were posted. When I refreshed the volume, I saw it turned to be 25% problem. Hmm... This means that big percentage might mean very hard problem which everyone tried to optimize at least to 3 seconds. Also it means that I can post some problem I solved numerous number of times to turn the hardest one into 99.9% if I will spend weeks optimizing it. Is that right? I always thought this percentage is num_ac_authors/num_submits, not just raw_num_ac/num_submits.

[shamim]

Yes, this is how it should have been done. Each author being counted once.
The suggestions given in Felix-Halim is also misleading due to this.
But there is a place (harvestsoft.com, i think ) that considers no. of author and not total AC.

[Destination Goa]

acm.timus.ru has its difficulty rating as num_ac_authors / num_attempted_authors

But I think that num_submits should be raw number to show if the problem "suggests" a lot of WA before one gets 1st AC. It's like a "tricky cases" rating.

It is doubtful question whether to count those raw_num_non_ac before 1st AC from the author. So that only actually stubborn solvers affect the final rating. Or at least not to count num_non_ac from the author unless he made at least 10 submits on this problem.

I.e. I suggest the formula:
% = num_ac_authors / sum(for each author, solved || num_non_ac_author >= 10 ? num_non_ac_author : 0)

[fh]

shamim wrote
Yes, this is how it should have been done. Each author being counted once.
The suggestions given in Felix-Halim is also misleading due to this.

That's true, it's more accurate to count the number of people got accepted on the problem, not the total number of accepted. The problem is, I cannot grab the data easily, since I'm not the admin of the uva site I can only grab information that is available on the site: the number of accepted.

If uva provides easy to access information about the number of (unique) people accepted for the each problem, then I can generate the ranklist. For now, the information should be grabbed from each problem statistics! this will bring the uva site down if i tried to grab that information often!

Solution: uva site should provide a kind of "xml data" to provide easy access on all of the information needed for me (and others) who wants to generate proper ranklist. Or... if they have spare time, why don't you asked them to revise/build the ranklist for that.. it would be much faster accessing database locally.

[misof]

acm.timus.ru has its difficulty rating as num_ac_authors / num_attempted_authors

But I think that num_submits should be raw number to show if the problem "suggests" a lot of WA before one gets 1st AC. It's like a "tricky cases" rating.

It is doubtful question whether to count those raw_num_non_ac before 1st AC from the author. So that only actually stubborn solvers affect the final rating. Or at least not to count num_non_ac from the author unless he made at least 10 submits on this problem.

I.e. I suggest the formula:
% = num_ac_authors / sum(for each author, solved || num_non_ac_author >= 10 ? num_non_ac_author : 0)

The difficulty rating is OK, perhaps along with num_ac_authors. A problem, which was attempted & solved by 3 people is far harder than a problem attempted by 1000, solved by 700.

I would say that the average number of submits before you get AC is a better "tricky cases" rating.