Hi everyone! #justforfun Ever wondered if the task you are trying to solve is really hard?
I'm happy to present tool that shows approximate rating of the problem. It needs to know only programming language of some accepted solution, its work time and used memory: https://ratingers.pythonanywhere.com/cf-task-rating
Autoselecting language should work, but if you're talking English and see my page in Russian, please tell me about that.
As tasks can be really different, accuracy is a bit small: average error is 450 rating.) I think it's a fundamental restriction, because there may be easy problems using more time than harder ones.
It's WIP. If you find any bugs, you can talk to me here!
I just solved problem D of last div2 on my own(hard for me) and then used your predictor to approximate the difficulty of this problem,the result is 1277.3.It seems like it is telling me:"Don't be happy so soon,you are still too weak!"XD
Yes, the accuracy now is small. (By the way, I can't see approximate rating of D in its tags.)
To get better accuracy, you can try taking some (3-5) accepted solutions and calculating average task rating based on them. Maybe, this will work better.)
got it : D
May be u have entered the memory in KB but its asking in bytes
Haha no,I used bytes.;)
Is this some sort of shitpost I'm too stupid to understand?
This is some case of programs written just for fun.
It says that 1667F has a rating of 1500 :)
I think, time and memory consumption is not appropriate input for such a prediction. It would be more accurate if using information about ratings of users who solved the problem or, for example, time for solving this problem.
Using rating of people who have solved the problem is a bit boring), AFAIK Codeforces uses these ratings itself.
About the time of solving problem: yes, that's an interesting idea, and I'll add it! But it will make the program harder to train, because now I train it on archive submissions.
You can use CF API and information about previous rounds (including problem submission time since contest start) for this purpose. There is quite convinient API wrapper for Python. It is simple enough to use.
Yeah, I'm already using CF API.
Problem submission time is too dependent on its position in the contest. Maybe, it will be better to use problem solving time: time from previous AC (for other task) to the current AC.
Updated the site: https://ratingers.pythonanywhere.com/cf-task-rating?v=1
Both old and new versions are available, but when using time of submission accuracy seems to reach 300 rating points!
Time, memory, language isnt the most important features of the complexity of the task. For example, some people haves identical idea implement differently. More important features can be count people who passed the task or tags problem. Also, I try to use your tool and 1394E - Boboniu and Banknote Collection have rate 1500.
The problem is, during contest you don't see task tags, and count of people solving it increases non-linearly. So during contest you'll be unable to see approximate rating of task.