Elo Rating System¶
Context¶
The Elo rating system is a mathematical method used to calculate the relative skill levels of players in zero-sum games like chess, created by Hungarian-American physics professor Arpad Elo in 1960. It was adopted by the International Chess Federation (FIDE) in 1970. Currently the system has been improved by platforms such as lichess and chess.com but we shall consider its simpler, original version.
Let’s see how it works. Consider the following temrs:
Score: The score of a chess match between a player and his/her oponent is either win (1 point), loss (0 points) or draw (0.5 points).
Elo It is a positive integer intended to reflect the level of a player relative to the players community.
Initial Elo: The organization provides a new player with a fixed initial elo amount.
Expected Score: The system calculates the probability of a win, loss, or draw based on the rating difference between two opponents. For example, a player rated 100 points higher is expected to win roughly 64% of their games. This probability is computed by the following formula where \(elo_{you}, elo_{oponent}\) are the player’s and the oponent’s elo ratings (the constant 400 value is a convention):
(1)¶\[E(elo_{you}, elo_{oponent}) = \frac {1} {1 + {10}^{\frac{elo_{oponent} - elo_{you}} {400}}}\]Elo Update: You gain points when you perform better than expected (e.g., beating a higher-rated player), you lose points when you perform worse than expected, and you neither gain nor loose points if you draw to a player rated like to you. The formula to compute that is as follows:
(2)¶\[elo_{new} = elo + k*(S-E(elo, elo_{oponent})\]where \(elo\) is the current elo, \(S\) is the actual score (win=1, draw=0.5, loss=0), \(E\) is the expected score and k is a constant factor that determines how much influence a single match has on your elo. It is decided by the organization. Since the result of this formula might have decimals but the elo rating used in reality is an integer value, the result is rounded to no decimals.
These formulae are needed to implement some of the requested functions but you don’t have to implement then since their respective Python functions are available in the elo_f.py file:
# Lluís Vila Grabulosa # ETSEIB # 13/Oct/2026. def exp_score(ra, rb): ''' INPUT: - ra, float, the player's elo rating. - rb, float, the oponent player's elo rating. OUTPUT: - The probablility or a win rounded to 2 decimals. REMARK: The 400 constant is a system convention. ''' return round( 1 / (1+10**((rb-ra)/400)), 2) def elo_update(elo, score, op_elo, k): ''' It updates the current rating of a player with *score* against a player of op_rating. It computes ''' return round(elo + k*(score - exp_score(elo, op_elo)))
Goal¶
Now, we wish to process the results of a number of players. The information in represented as a Players List where for each player we have the name, current elo and a list of results (scores) from a number of matches (not necessarily the same number for each player). See an example below:
>>> l = [ ... ['Magno', 1800, [(1.0, 1900), (1.0, 1960), (1.0, 2020)]], ... ['Fabi', 1800, [(1.0, 1900), (0.0, 1900), (1.0, 1900), (0.5, 1920)]], ... ['Ali', 1800, [(1.0, 1800), (0.0, 1900), (0.0, 1820), (0.0, 1750)]], ... ['Sinda', 1800, [(0.0, 2000), (1.0, 1800), (1.0, 1900), (1.0, 2000)]], ... ['Nyman', 1800, [(0.0, 1800), (0.0, 1750), (0.5, 1700), (1.0, 1780)]], ... ]
Observe that the Scores List is a list of tuple. Each tuple has two components:
Bearing these data structures in mind, you are required to implement the following functions and deliver them in the module elo (file elo.py). The weights of these functions are 10%, 50% and 40% respectively.
Warning
Do not leave pieces of code that are wither incomplete or contain syntax errors in the .py file to deliver since that would result in a 0.0 grade for the whole problem.
Auxiliar Function¶
This is an auxiliar funcion that will be useful to implement the main functions later on. It is intended to compute the new elo of a player that played a number of matches. Hence, the new elo is calculated from the current elo and the score list of those matches:
- elo_calculate(ini_elo, scoreL, k)¶
such that
given
returns a
intwith the new elo calculated from the current elo, the scores in scoreL sequentially in that order, using the given k factor and applying the Elo Update formula above (2).
For example:
Doctests are available in the elo_calculate.test file.
Main Function #1¶
This function is intended to separate the players according to the change in their elo rating: positive, equal or negative change:
- elo_delta_classify(playersL, k)¶
such that
given
returns 3 new
listof players summarized info:
The list of the players whose elo increased. The list must be ordered decreasingly.
The list of the players whose elo change is less than 10 points. The list must be ordered increasingly.
The list of the players whose elo decreased. The list must be ordered increasingly.
All are
listoftuplewhere each tuple contains the following players’s information:The input list should not be modified.
For example:
>>> l = [ ... ['Magno', 1800, [(1.0, 1900), (1.0, 1960), (1.0, 2020)]], ... ['Fabi', 1800, [(1.0, 1900), (0.0, 1900), (1.0, 1900), (0.5, 1920)]], ... ['Ali', 1800, [(1.0, 1800), (0.0, 1900), (0.0, 1820), (0.0, 1750)]], ... ['Sinda', 1800, [(0.0, 2000), (1.0, 1800), (1.0, 1900), (1.0, 2000)]], ... ['Nyman', 1800, [(0.0, 1800), (0.0, 1750), (0.5, 1700), (1.0, 1780)]], ... ] >>> l1 = [(1866, 'Magno', 1800), (1851, 'Sinda', 1800), (1832, 'Fabi', 1800)] >>> l2 = [] >>> l3 = [(1771, 'Ali', 1800), (1780, 'Nyman', 1800)] >>> (l1, l2, l3) == elo_delta_classify(l, 32) True
Note
Notice that in this example the second list is empty since all players either increased or decrease their rating.
In your implementation of this functions it is recommended to call the previous function.
Doctests are available in the elo_delta_classify.test file.
Main Function #2¶
This is function intended to add scores to a player in the players list:
- elo_player_extend(playersL, name, scoreL, k)¶
given
modifies playersL by adding the scores in scoreL associated to the player called name.
returns and
intthe new elo rating for the player called name calculated from the extended scores list. If the name is not found in the players list then the values returned will be -1.
For example:
>>> L = [ ... ['Magno', 1800, [(1.0, 1900), (1.0, 1960), (1.0, 2020)]], ... ['Fabi', 1800, [(1.0, 1900), (0.0, 1900), (1.0, 1900), (0.5, 1920)]], ... ['Ali', 1800, [(1.0, 1800), (0.0, 1900), (0.0, 1820), (0.0, 1750)]], ... ['Sinda', 1800, [(0.0, 2000), (1.0, 1800), (1.0, 1900), (1.0, 2000)]], ... ['Nyman', 1800, [(0.0, 1800), (0.0, 1750), (0.5, 1700), (1.0, 1780)]], ... ] >>> Lcopy = L.copy() >>> res = elo_player_extend(L, 'Pepe', [(1.0, 2200), (0.5, 2700), (1.0, 2580)], 32) >>> res == -1 and L == Lcopy True >>> res = elo_player_extend(L, 'Sinda', [(1.0, 2200), (0.5, 2700), (1.0, 2580)], 32) >>> res == 1926 and L[3] == ['Sinda', 1800, [(0.0, 2000), (1.0, 1800), (1.0, 1900), (1.0, 2000), (1.0, 2200), (0.5, 2700), (1.0, 2580)]] True >>> res = elo_player_extend(L, 'Magno', [(1.0, 2800), (0.5, 2850)], 32) >>> res == 1914 and L[0] == ['Magno', 1800, [(1.0, 1900), (1.0, 1960), (1.0, 2020), (1.0, 2800), (0.5, 2850)]] True
Doctests are available in the elo_player_extend.test file.