The 19 parameters include separate \(K\) factors for Online, Offline and Tier-1 events. Higher tier events, higher tier players and LAN events get larger weight in the scoring function. This setting (and the 17 others used in my Elo algorithm) are set by optimizing the settings to produce good probabilities and accurately predict matches. One change that improved my results: when you switch games, you get 70% of your current Elo (across all games) and 30% of your last Elo in that game (which is the default 1,500 if you’ve never played it before). In total, I’ve about 17,000 1v1 matches.įor ArenaFPS I found that combining ratings across games a significantly better log likelihood model, so it’s what I’m using. In this case, I added an extra 70 tournaments of match data to the wiki myself, just to patch up some holes. The data is imperfect, with missing events and likely a few errors, but it’ll have to do. I download an export of Liquipedia and build an Elo rating system from the historical data. The methods here are the same as in my previous writeup. Following my previous post analyzing Counter-Strike, here I decide to answer who were the greatest 1v1 FPS players of all-time? With COVID-19 continuing the shutdown of sports leagues, the itch to do sports analytics leaves us with eSports.
0 kommentar(er)
