watti

CS2 scouting service SAURON Battle tested. DMs open - 1 million high elo demos - Statistically proven detection of undervalued ELO players who get in 4k - Shortlist builder - Voice analysis: IGL, tilt, qualtity - HLTV3 - Age prediction - Aim metrics: honest vision + AnimGraph2

@smartbackwards what I noticed about Boombl4 high elo matches is that besides not a shiny K/D and stuff the round swing model shows him as a very valuable player

close but not cigar for big impact boombl4 he had 6 rounds over 25% of my rd.swing on that nuke, ZywOo 30 bombs with just one 💀 my swing has wOo at +5.13% for the map, im almost matching HLTV on CT but on T sides ive got him as a +3.1% instead of +5.45% #biased

my internal path builder tool. foundation for the tactics evaluator used for: - where could opponent physically be in N sec - real travel distance - save/hero play detection accounts for the jumps and boost as shown in the. so not just a regular bots .nav

@smartbackwards proper approach with utilizing prediction function. wonder what are your model performance metrics the most interesting to beat are ~50/50 situations - overall 0.99 auc is trivial I will def post the one I am using later noticed how it improves with more game state features

we're saying goodbye to PARIVISION, but hello to a project i've had on the backburner for a while - map dashboards it features some stats ive developed: the map win probability chart, xR, EEK and, as a little treat, something new - impact rounds (name wip), number of rounds in which according to my round swing calculation a player had over a 25% impact on the round being won it should tell you everything that happened in a given map at a glance - here we see that PARI had a big lead and despite shaky stuff from legacy, dumau and latto were enough to pull the comeback over the finish line. looking for feedback! wont post these for every map (maybe for the playoffs), but let me know if you even like this or not

@ChainerNotes dm me

@n3c1k you have some discord or whatever? i am building an analytical tool too, thought we might share some insights or just generally useful stuff. been grinding my head around various positioning valuation metrics for a while now.

smoooooth animgraph2 external replay that's my internal tool used to get precise aim related stats for the statistics for those who wonder how the real model looks vs hitbox - its shown a bit in the video might consider adding textures and lightning for fun and exercise

@ChainerNotes ah, no its not like I do picture rendering on GPU and then hook it, would be much slower you still process map geometry as u said, just with GPU instead of CPU, that's what is meant by "external gpu" in the community

@n3c1k what i meant by LOS raycast is using tri static map mesh + raycasts in order to determine visibility of players. u previously said u use "GPU vision detection with animation reconstruction", so i wondered how much more accurate is that compared to tri mesh+raycasts

@ChainerNotes maybe I dont understand what is GPU rendering method u mean

@n3c1k thanks, got it. I also once asked a question under your other post, but i'll ask here too: have you compared the actually accuracy of GPU rendering method vs LOS raycast? Since rebuilding a scene from demo's ticks is quite heavy, i wondered if it yields substantial improvement

@ChainerNotes sorry for missing your previous question - always happy to get one yes, I do have automatic quality controls in basically every program in this case I use model rendering memory dump as an oracle, then whatever I do has to exactly match it on each tick

@ChainerNotes yeah that's just browser rendering performance, the animation tickrate is bound to 64

@n3c1k you still base it on the demo's tickrate, right? i am just trying to understand where additional FPS comes from if demo essentially provides us with 64 ticks per second Or is it purely 3d scene recreation fps render?

you know sometimes you make a strafe and it feels sluggish? its figured out now, happens after you make a big clutch

@Aluminaticsgo short answer is proven by a prediction model with a proper histoeical backtests from my experience, using the same approach I have for 4000 elo gives noticeably better performance in a lower bracket (weights would differ tho) didn't dig much into it, just noticed the effect

@n3c1k If you were to predict the next amateur players to breakout, what stats would you use?

scouting is an all time task in sports lack of data is a common issue in traditional sports, CS is digitalized by its nature its not a question anymore if u can predict elo breakout players I do expect more people to go for it. the question would be who's model is doing better

some one messaged me in discord saying that my work inspires them to make something reality is when they go for it, data providers will actively try to not let them get the demos will this guy spend 1 year studying "data collection techniques" or just pursue into other field?

when you talk to the people from the data community or teams, everyone says: we need this data, but they dont give us. no one is happy with this things could be so much better Counter-Strike has a lead over traditional sports, and providers dont let the community to use it

unfortunately, demos data gatekeeping is a thing and something has to be changed data providers are heavily reducing community ability to access demos I've got what I need. I have a working model with 60-95% precision and recall this problem slows down the CS analytics field

@footknockout yeah I am going to make it publicly accessible soon free? that's not how the things work

@n3c1k Will this go public at some point, and if so, would it be free? This seems really useful for my 10k premier lobbies

next thing I plan to announce is a grenades predictor just a small teaser for now - need to prepare the post - molotov burn area prediction - 1 billion pre calculated grenades - web interface 10 million trajectories per second in live more accurate than the game sometimes :)

@divinorumA teams prefer to use training tools instead of cheats, if some1 didnt offer it for 10 years its their personal problem if you read my message again, I have 1 billion precalculated grenades. show me cheat provider with such amount ah, they dont need it? exactly, different goals

@n3c1k Cheats have had ts for like a decade atp and they’re far smoother than whatever this is 😭

@Huwjr you have to consider this very seriously, and it happens sometimes funny story: I was in a situation when I had some HLTV statistics that clearly exposed some cheaters, but didn't report to not disclose the statistical features of the client

@n3c1k p.s i am impressed by what you've made, this is not me crap talking your tool. genuinely think some of these stats would help faceit detect more accounts but they probably dont care.

some content managers would say: don't post during the game day evening, what are you doing well I don't have managers, so I made the post anyway slowly crawling into the main page, lets gooo reddit.com/r/GlobalOffens…

@SamCarterCS I have all games processed for every region. EU is being dominant with like x30-x50 more players in the ladder, so I've got all the data If there would be an interested organization I would be happy to dig into it and uncover some talent the data is on my SSD, waiting for it

@n3c1k Any coverage of the Asian/oceanic region in terms of the scope of the project at the moment? I’m sure you might now how unloved our regions are sometimes are in terms of the esports scene or even trying to get noticed as talent