There was a part I had in an initial draft (that I eventually cut out) where I used the idea to "predict" the round wins of the teams based on those stats, and it's possible that metric might be more predictive of a team's future performance than actual round wins but that wasn't something I looked into for the article.
Projections are definitely something I'm most interested in doing in CS long term though, especially when it comes to individual players moving between teams/regions/tiers, but that's a long way off for me before I even know where to begin with all that.
yeah I don't love using rounds to predict win probability. It's basically using a metric to predict itself (because rounds = wins). The best sports models for example *don't* use points to predict future wins, which is counterintuitive but more accurate.
If I had all the raw data in the world, I could probably come up with something more predictive than HLTV 2.0 ranking, but as a shortcut it works out pretty well.
Small chance you’ll see this but I’ll give it a shot!
How are you pulling the raw data? Did you go through manually or is there an easy way to do it? I’d like to play around with some of the updated numbers myself.
Separately, one thing to note is that Leetify has an “aim rating” that weighs things differently. I haven’t been able to find the exact weights but spotted accuracy seems to be the most important factor from my tests.
You need to go to each individual player's Leetify page using the "Spectate" feature and adjust the ranges and change the data source to "HLTV". You don't get aim rating this way, just the raw numbers.
To get all the aim data, I did have to go through and collect every player's numbers manually. I'll be doing an updated article at the end of the year with numbers for all of 2024.
It's not 100% accurate because there's some rounding going on which we can't see since Leetify presents only rounded numbers.
But I DO think this is more accurate than just averaging each percentile like you did. Some things are just more important (i.e. spotted accuracy vs. counter strafe).
My top 10 (based on my formula + your raw data) RIFLER only ends up being:
1. ropz
2. NiKo
3. KRiMZ
4. misutaaa
5. ISSAA
6. TRAVIS
7. zont1x
8. skullz
9. donk
10. d4rty
This is rifler only...you should never compare ratings of AWPers and Riflers - one of the many reasons being Leetify bizarrely doesn't differentiate the crosshair accuracy between the two - and AWPers almost always have way lower crosshair accuracy degree just given the nature of the gun.
I think even ratings of AWPers compared to each other are garbage because how often they pick up the AWP makes a huge difference. Easy fix by Leetify would just be not to count any ratings when any player has an awp (or anything that scopes) - only count rifle ratings (the "main 4" - AK / M4 / Galil / Famas).
(Note...to add ANOTHER wrinkle, the aim ratings on the specific match pages aren't the same as the aim ratings on the player page - see Ropz for example, he has a 98 overall rating but his individual matches are more in the 80's / low 90s - the overall rating seems to be a percentile maybe? I have no idea.)
I don't mind looking at aim rating but the title of the article is more of an attention grabber, I just wanted to examine the characteristics of all the players and see how they compared. The average percentiles aren't meant to be used as a true comparison, it's just a rough guide.
Sorry - one last question - do you have the leetify URLs of all the top players saved down? I find it hard to find all the players since so many people copy their names. I just start with ropz then go through his recent matches - but it's a grind...
It probably would have been smart of me to do that, but nope. If you go to each player's Liquipedia page you can find their Steam URL and put that into the Leetify search instead of their in game name. If the Steam URL on Liquipedia doesn't work you can do the same thing on Faceit, which is usually also linked on Liquipedia.
BTW - you might like this thread I posted on HLTV a couple of weeks ago. Good to see someone else interested in CS analytics.
https://www.hltv.org/forums/threads/2912679/fixed-2024-hltv-team-rankings-using-computer-model
This is definitely interesting, this falls more in line with my most recent article about value and whatnot. https://nomiun.substack.com/p/how-much-is-a-kill-worth
There was a part I had in an initial draft (that I eventually cut out) where I used the idea to "predict" the round wins of the teams based on those stats, and it's possible that metric might be more predictive of a team's future performance than actual round wins but that wasn't something I looked into for the article.
Projections are definitely something I'm most interested in doing in CS long term though, especially when it comes to individual players moving between teams/regions/tiers, but that's a long way off for me before I even know where to begin with all that.
yeah I don't love using rounds to predict win probability. It's basically using a metric to predict itself (because rounds = wins). The best sports models for example *don't* use points to predict future wins, which is counterintuitive but more accurate.
If I had all the raw data in the world, I could probably come up with something more predictive than HLTV 2.0 ranking, but as a shortcut it works out pretty well.
Small chance you’ll see this but I’ll give it a shot!
How are you pulling the raw data? Did you go through manually or is there an easy way to do it? I’d like to play around with some of the updated numbers myself.
Separately, one thing to note is that Leetify has an “aim rating” that weighs things differently. I haven’t been able to find the exact weights but spotted accuracy seems to be the most important factor from my tests.
You need to go to each individual player's Leetify page using the "Spectate" feature and adjust the ranges and change the data source to "HLTV". You don't get aim rating this way, just the raw numbers.
To get all the aim data, I did have to go through and collect every player's numbers manually. I'll be doing an updated article at the end of the year with numbers for all of 2024.
Ooooof - I was worried that was the case.
I did some regression analysis to try and back into the weights on the aim rating. I got:
= 4.77 + (84.34 * Spotted Accuracy) + (-0.05 * Time to Damage) + (-1.65 * Crosshair Accuracy) + (74.40 * Headshot Accuracy) + (38.06 * Spray Accuracy) + (56.55 * Counter Strafe)
It's not 100% accurate because there's some rounding going on which we can't see since Leetify presents only rounded numbers.
But I DO think this is more accurate than just averaging each percentile like you did. Some things are just more important (i.e. spotted accuracy vs. counter strafe).
My top 10 (based on my formula + your raw data) RIFLER only ends up being:
1. ropz
2. NiKo
3. KRiMZ
4. misutaaa
5. ISSAA
6. TRAVIS
7. zont1x
8. skullz
9. donk
10. d4rty
This is rifler only...you should never compare ratings of AWPers and Riflers - one of the many reasons being Leetify bizarrely doesn't differentiate the crosshair accuracy between the two - and AWPers almost always have way lower crosshair accuracy degree just given the nature of the gun.
I think even ratings of AWPers compared to each other are garbage because how often they pick up the AWP makes a huge difference. Easy fix by Leetify would just be not to count any ratings when any player has an awp (or anything that scopes) - only count rifle ratings (the "main 4" - AK / M4 / Galil / Famas).
(Note...to add ANOTHER wrinkle, the aim ratings on the specific match pages aren't the same as the aim ratings on the player page - see Ropz for example, he has a 98 overall rating but his individual matches are more in the 80's / low 90s - the overall rating seems to be a percentile maybe? I have no idea.)
I don't mind looking at aim rating but the title of the article is more of an attention grabber, I just wanted to examine the characteristics of all the players and see how they compared. The average percentiles aren't meant to be used as a true comparison, it's just a rough guide.
Sorry - one last question - do you have the leetify URLs of all the top players saved down? I find it hard to find all the players since so many people copy their names. I just start with ropz then go through his recent matches - but it's a grind...
It probably would have been smart of me to do that, but nope. If you go to each player's Liquipedia page you can find their Steam URL and put that into the Leetify search instead of their in game name. If the Steam URL on Liquipedia doesn't work you can do the same thing on Faceit, which is usually also linked on Liquipedia.