Some time ago I posted this blog, which showed the diminishing effect of a huge rating loss over a period of many contests. Recently, I decided to try to show this effect a bit more broadly. So I extended it to be a bit more extreme and simulated my entire contest history while ignoring all rating losses. That is, I ran through every contest I was ever rated for and calculated the expected rating change with cfviz, but maxed out every rating change with $$$0$$$.
Now, the question is, how much of an effect did ignoring rating losses have? I give more details in this video, at the 6:56 mark (said video also has many more tips, generally being about how you can learn more from the contests you do, which is not-so-subtly the whole point of this blog). But, essentially...
the "optimistic" rating was $$$2662$$$, only $$$50$$$ more than my current rating. There's probably some sort of bias here, one of the possible issues being that I've had an unnatural streak of rating rises within my last few contests, which would likely significantly decrease the difference. You can find the full data here, which probably does show that $$$50$$$ is kind of below normal. Taking the average over all differences gives $$$90$$$, which is maybe more accurate. You can decide for yourself how significant this number is, but I would say that it's not that large of a difference, given that it's as optimistic as it possibly could be.
great video! I will definitely do more contests and upsolve to improve!
Rating doesn't actually equals to ability... So it probably matters less.
Define "ability". Honestly, I think rating is a pretty good measure.
Isn't rating very shaky? At least it's for me. Idk but rating is probably the best measurement we can get.
I think it still works pretty nicely (maybe it's better to look at the general trend instead of the current exact value). Also, AtCoder rating system is less shaky if you prefer that.
I believe that, ability, is the best performance you can ever have. hence, your highest rating could probably be a better measure of your ability.
So if you only practice tree dp, can't do anything else, and a contest with all tree dp problems comes along and you get first place, that is a measure of your ability?
Actually, that shows your ability on tree dp problems.
And by the way, what is your definition for ability?
Ability is about consistent performance, not your best performance ever, which for cp means you need to be well versed in all topics. Each individual contest doesn't mean much in my opinion, but as they average out over time in your rating that shows your true ability. If you get lucky and win the lottery once, do you have more ability than others at winning it?
Actually in the lottery case, I believe it’s more about the probability of winning, i think we can say that probability of winning is equal to your ability in winning, but i don’t think this is like math probabilities in which we can say it’s 1 / (the # of participants). I’m not sure about it’s definition here...
The analogy is a single contest you can just get lucky on still with topics you know very well, getting everything bug free first time if it's unusual for you, or even having seen a problem essentially the same before. Sure it's not the same as a lottery, but the point is a single contest still relies on luck. And that's the point of this blog in the first place, but rather than luck it is instead speaking of unlucky contests will be averaged out to your true skill.
Like, how well would you perform developing a implemental programme or participating contests like ICPC. It doesn't seem objective either, though. So don't mind it, I was being dumb :(
It gives me 102. Probably because I was super inconsistent in contests. Also I often need to take break for some various reasons, currently I'm busy with my school activities so can't participate in contests.
What's your advice on what to do when someone would return? Doing some virtual contests before participating or straight going into contests?
LOL not if you lost 200 on the last contest