Summary of The Reasonable Ineffectiveness of Math in League

00:00:00 - 00:30:00

The video discusses the ineffectiveness of math in predicting outcomes in the game League of Legends. It argues that complex models are not always better in predicting outcomes, and that in some cases they are even harmful. The video also provides a link to an article describing how stats in baseball measure a player's individual skill, which is not possible in league.

• 00:00:00 The author critiques math being useless in League, pointing out that complicated models don't provide anything useful beyond the obvious. They then discuss how a model can be useful in separating players into roles and estimating their medium kda. The author finds that support and jungle tend to have high kda and win rates, and that the model is accurate to 85 percent for those roles.
• 00:05:00 The author creates a 10 gold opponent column to measure if you win lane by comparing your goal at 10 minutes to your opponent's goal to 10 minutes. This "stat" is only somewhat correlated with winning the game. The author finds that player who won lane won 60 of the games, but that's not close to the 82 percent we're going for. By comparing your goal to your counterpart at the end of the game, the author finds that this one stat is even more accurate. The reason for this is that the stats that perform best here are the ones that are most correlated with winning, not necessarily the ones that best isolate your individual performance. When you use gold earned pike is like a totally different champion than everyone else, because his old just gives like thousands of gold bonus gold like every game if he you know executes people with like the right way to think of pike as a champion is like he's kind of useless all he can do is kill people. Like he gives your team like a giant gold lead for killing people. So by chopping off like a 2,000 gold pike tax if you play pike or give you like a 2,000 gold standard deduction, the author is able to predict 4 out of the 5000 games.
• 00:10:00 The video discusses how math models used in baseball can be ineffective, as they do not account for player performance. The author also argues that complex models are not always better in predicting outcomes, and that in some cases they are even harmful.
• 00:15:00 The video discusses the ineffectiveness of math in sports, specifically in baseball. It argues that stats such as batting average and runs batted in are not as indicative of a player's performance as they are in other sports, and that advanced stats such as on base percentage are more important in league. The video also points out that although league is more complex than baseball, it is still possible to understand a player's performance by watching them play. Finally, the video provides a link to an article describing how stats in baseball measure a player's individual skill, which is not possible in league.
• 00:20:00 The article discusses how baseball fans suddenly learned to do math and how that has led to better data collection and analysis. It goes on to say that, while individual player stats can be valuable, it is more important to focus on team stats. OpenAI, a company that has created an agent to play Dota 2, has developed a model that learns how to maximize rewards by looking at game data.
• 00:25:00 In "The Reasonable Ineffectiveness of Math in League," the YouTube author discusses how math may not be the most effective tool when it comes to playing League of Legends competitively. One of the challenges that was brought up was that the game can change significantly with each patch, making it difficult for even the best players to keep up. Furthermore, the author explains that OpenAI's own self-training model is able to learn to play different games based on their own specific set of rules, even if those games are from previous patches that the model has not seen before. Finally, the author discusses how the model could be used to measure human player performance in a similar way to how a computer can determine if a chess decision is good or bad.
• 00:30:00 The author of the video expresses disappointment at the quality of some of the comments on a post he made, specifically those which seem to be promoting a product or service. He suggests that users be careful when commenting on posts that they are affiliated with any companies or groups, as this can be dishonest and manipulative. He also suggests that users create a common data set of information for use in projects predicting the winners of leagues or other competitions.