DeadMan69's blog

ML Based Rating Predictor
By DeadMan69, history, 2 days ago, In English

Hello Codeforces,

This is my first blog on Codeforces, quite excited :) (hopefully this won't be the last).

I wanted to share this ML based application that I created, which predicts a user's rating using Linear Regression, based on the following factors:

  • Number of Problems Solved
  • Average Rating of Problems Solved
  • Registration Date

Application Link : CF-Rating-Predictor
Github Link : Github-Repo

This is my first time actually implementing any thing ML Related, sorry in advance in case I wrote something wrong ;-;

Thank you for your Time ^_^. If you have any suggestions for improvements, do let me know :)

  • Vote: I like it
  • +81
  • Vote: I do not like it

»
2 days ago, # |
  Vote: I like it +17 Vote: I do not like it

It estimates me at 1434 lmao.

→ Reply
  • »
    »
    2 days ago, # ^ |
      Vote: I like it +1 Vote: I do not like it

    It's a bit Inaccurate, when the growth of the user is exponential like in your case or for people with 3000+ rating, will try to make it better by adding more variables and making the dataset larger, currently it's only 5000 users (took me 5 hours to just load 4000 user's data from CF :/ ), let's see how much more accurate I can make it.

    → Reply
»
2 days ago, # |
← Rev. 2   Vote: I like it +5 Vote: I do not like it

it's so over

But I'm kinda confused as to what this is predicting. Is it trained on stuff?

→ Reply
  • »
    »
    2 days ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Yeah , trained it on 5000 random users from Codeforces, tried to avoid all the id's which could have potentially been alt and took the variables x = [number of problems solved, avg rating of problem solved, registartion date] and y as the rating of the user, after this , based on the [number of problems solved, avg rating of problem solved, registartion date] in the input, it predicts the rating of the user, in your case, you have solved plently of high rated problems, so it went overboard a bit XD.

    → Reply
    • »
      »
      »
      2 days ago, # ^ |
        Vote: I like it +5 Vote: I do not like it

      Interesting, it seems to be pretty accurate in a lot of cases ($$$\pm 150$$$ or so). Though it gave you an estimated rating in the $1500$s. Very cool stuff.

      → Reply
      • »
        »
        »
        »
        2 days ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        Yeah, I was also disappointed with the ratinng it predicted for me lmao , that just means I need to solve higher rated problem XD

        → Reply
        • »
          »
          »
          »
          »
          2 days ago, # ^ |
            Vote: I like it 0 Vote: I do not like it

          I actually wonder if it says something about $$$IQ$$$. Like users with lower predicted rating than true rating have higher $$$IQ$$$ and vice versa. Cuz they were able to do worse/better with the same amount of problems solved. The only other thing I've found that has something like this is the graph from this study (which is probably better at predicting $$$IQ$$$ than just flat out rating), but yours accounts for difficulty, so it could potentially be better.

          → Reply
          • »
            »
            »
            »
            »
            »
            2 days ago, # ^ |
              Vote: I like it +8 Vote: I do not like it

            Yeah could be, but there is also the fact that some people use other resources for practice, like 1-2 month ago, I started doing CSES sheet (pretty dope sheet in my opinion, give it a try if you haven't).

            I was also thinking of doing something like using clustering to create subgroups based on some factors that could be related to IQ, and then perform the regression for people having similar IQ, this way it would have been a bit more accurate in my opinion, let's see will try this as well if I get the time :) .

            → Reply
»
2 days ago, # |
  Vote: I like it +5 Vote: I do not like it

Somehow got a 1637 expected rating. Let’s hope I can reach expert this year :)

→ Reply
»
2 days ago, # |
  Vote: I like it +8 Vote: I do not like it

Can you try different models besides linear regression? Would be interested in which model is the most accurate.

→ Reply
  • »
    »
    2 days ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Yeah will do, will have to learn a bit more ML first XD, but will certainly post a blog again once I do :)

    → Reply
»
2 days ago, # |
  Vote: I like it +5 Vote: I do not like it

For my friend, who reached master, the predictor gives the expected rating of 1573. Was he really that lucky lol?

→ Reply
»
2 days ago, # |
  Vote: I like it 0 Vote: I do not like it

I did not just get called incompetent by an AI (again)

Jokes aside cool project

→ Reply
»
2 days ago, # |
  Vote: I like it 0 Vote: I do not like it

Sadly I think it's not that great:(, for tourist it's showing expected 3100 around, even though he got many 1-st ranks.

→ Reply
  • »
    »
    43 hours ago, # ^ |
      Vote: I like it +3 Vote: I do not like it

    Yeah well, people with 3000+ rating are too good to be predicted by AIs :(

    → Reply
»
47 hours ago, # |
  Vote: I like it 0 Vote: I do not like it

It predicts me at 2017, but I think partly because it takes the registration date into account. It's better to take the first rated contest date instead of the registration date imo. I started doing contests almost 3 years after registering, so those 3 years means nothing.

I also think it should take the recent contest performances as well. Interested to see how it performs for someone with weird rating graphs like mine.

→ Reply
  • »
    »
    43 hours ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Yeah that's a great idea , I think it might be better to take number of active days into account, since people take breaks and all, will try this :)

    → Reply
»
47 hours ago, # |
  Vote: I like it 0 Vote: I do not like it

It says my expected rating will be only 2132. It's interesting that I've never reach ~2100 rating before.

→ Reply
»
47 hours ago, # |
  Vote: I like it 0 Vote: I do not like it

Could it estimate future rating? You could use data as of x months ago as predictors and current rating as the target

→ Reply
  • »
    »
    43 hours ago, # ^ |
      Vote: I like it +8 Vote: I do not like it

    Yeah, will try that, I even remember a blog where a guy would manually predict everyone's future rating and he ended up being quite accurate XD

    → Reply
»
45 hours ago, # |
  Vote: I like it 0 Vote: I do not like it

If my max is 1648 obviously someday my rating will be 1680 , why do I need a predictor for that?

→ Reply
»
45 hours ago, # |
  Vote: I like it -10 Vote: I do not like it

Propaganda

→ Reply
»
44 hours ago, # |
  Vote: I like it 0 Vote: I do not like it

How do you pick the factors?

→ Reply
  • »
    »
    43 hours ago, # ^ |
      Vote: I like it +8 Vote: I do not like it

    I created a graph for all the possible factors and the 3 factors (number of problem solved,avg rating of problem solved,registration date) had the most linear relation with rating, so I chose them. Before this I had tried with only the number of problems solved , and there was no realtion at all between rating and number of problem solved, in a way proving that quality of problem matters over quantity.

    → Reply
»
43 hours ago, # |
← Rev. 3   Vote: I like it -22 Vote: I do not like it

I don't see the point of predicting someone's expected rating, like

Does it motivate someone to grind harder? No

Does it give them insights on how to improve? No

Can they draw concrete conclusions from your output? No

→ Reply
  • »
    »
    43 hours ago, # ^ |
      Vote: I like it +9 Vote: I do not like it

    Upto now , I have just made this for fun, found it quite intriguing that's why shared, I could try adding insight, one could be something like telling them what rating problems to solve to prpgress, will try this as well, thank you for your input :)

    → Reply
»
42 hours ago, # |
  Vote: I like it +5 Vote: I do not like it

Sorry guys, I should have been 1700+

→ Reply