Remember the scene from the Facebook movie, The Social Network where Mark Zuckerberg, while creating “Facemash” (the earlier version of Facebook), tells Eduardo Saverin, "I need the algorithm you used to rank chess players” who then writes an equation on the glass panel?
That equation is called the Elo rating or Elo rating system, named after its creator Arpad Elo. This rating system is used to rate the skills of the players in competitor-versus-competitor games like chess, football, baseball, and American football.
Elo believed in the following:
- Performance of each player in a game is a normal distribution of random variables
- The mean value of players irrespective of their performance in an individual game increases slowly.
Initially invented as a rating system for chess players, Elo is now used as a fundamental rating system in most video games, snooker, scrabble, etc.
Elo rating system explained -
This system is used to determine the output of a game by using a player’s Elo rating. It is all based on probability. Players with a higher Elo rating have a higher probability of winning a game than players with a lower Elo rating.
After the game, the winner takes points from the loser, thereby increasing his rating.
If a high-rated player wins, only a few points will be transferred from the lower rated player. However, if the lower rated player pulls off an upset win, then the number of points that are taken by the player with the lower rating is far greater. (Sounds unfair, doesn’t it?)
Elo rating system equation explanation -
Let's take a look at the Elo rating equation.
calc = (1.0 / (1.0 + pow(10, ((rating_2 - rating_1) / 400))));
In this equation, RA and RB stand for the current Elo ratings of a player.
In real-world competitive gaming, a player has a probability of winning, losing, and drawing a match. So when a player has a score of 0.64, the probability of winning, losing, and drawing is 64%, 26%, and 0%, respectively.
After the match, if a player’s match score exceeds his predicted score, then the player’s Elo rating is updated. Similarly, when a player’s match score falls short of the expected score, then the player’s rating is adjusted downward.
Assume that a player is required to score EA during a tournament but scores SA, then the player’s rating is updated by using the following formula:
calc = (rating + kfactor * (actual - expected));
K is known as "K factor". It is a measure of how strongly a match will influence a player rating. If K is of a lower value, then the rating is changed by a small fraction but if K is of a higher value, then the changes in the rating are significant. Different organizations use different K factors; there is no universal value defined for it.
Pseudo code for the Elo rating system
import math; class elo_core: def getExpectation(rating_1, rating_2): calc = (1.0 / (1.0 + pow(10, ((rating_2 - rating_1) / 400)))); return calc; def modifyRating(rating, expected, actual, kfactor): calc = (rating + kfactor * (actual - expected)); return calc;
Elo rating system example -
Assume that Garry Kasparov has a rating of 2500 and Vishwanathan Anand has a rating of 2200.
Their expected scores are:
- If Garry Kasparov (wins) = 1 / (1 + 10 ^ ((2200 - 2500)/400)) = 0.849
- If Vishwanathan Anand (wins) = 1 / (1 + 10 ^ ((2500 - 2200)/400)) = 0.151
Now believe that the Federation has decided that the value of K is 24.
The new ratings of the players are as follows:
If Garry Kasparov wins- Garry Kasparov = 2500 + 24 * (1 - 0.849) = 2503
- Vishwanathan Anand = 2200 + 24 * (0 - 0.151) = 2196
- Garry Kasparov = 2500 + 24 * (0 - 0.849) = 2479
- Vishwanathan Anand = 2200 + 24 * (1 - 0.151) = 2220
Since Garry Kasparov (with a better rating) is the favorite, his win did not change the rating drastically. However, if the underdog, who in this case is Vishwanathan Anand wins, the ratings change drastically.
This is why in a cricket series between India and Zimbabwe, when Zimbabwe (with a lower rating) beats India, its rating changes quickly and it moves up the table. However, if India wins there is hardly any change in its ranking (Though ICC uses a modified
Let’s take the example of Elihu Feustel, who makes almost a million dollars a day by betting on tennis. You would think that he would recognize Serena Williams if he passed by her on the street. However, this isn’t the case, he neither recognizes any of the famous tennis players nor does he know their playing styles etc. So, just how does he know who to bet on? He relies on a big data model that the live betting market uses. This algorithm collects data from existing (almost 260,000) matches. This data is then manipulated to allow users to determine which player/match they should bet on.
