Functions to compute rating and ranking using Elo method.

rate_elo(cr_data, K = 30, ksi = 400, initial_ratings = 0)

rank_elo(cr_data, K = 30, ksi = 400, initial_ratings = 0,
  keep_rating = FALSE, ties = c("average", "first", "last", "random", "max",
  "min"), round_digits = 7)

add_elo_ratings(cr_data, K = 30, ksi = 400, initial_ratings = 0)

elo(rating1, score1, rating2, score2, K = 30, ksi = 400)

Arguments

cr_data

Competition results in format ready for as_longcr().

K

K-factor for Elo formula.

ksi

Normalization coefficient for Elo formula.

initial_ratings

Initial ratings (see Iterative ratings).

keep_rating

Whether to keep rating column in ranking output.

ties

Value for ties in round_rank().

round_digits

Value for round_digits in round_rank().

rating1

Rating of player1 before the game.

score1

Score of player1 in the game.

rating2

Rating of player2 before the game.

score2

Score of player2 in the game.

Value

rate_elo() returns a tibble with columns player (player identifier) and rating_elo (Elo ratings, based on row order, by the end of competition results). Bigger value indicates better player performance.

rank_elo() returns a tibble with columns player, rating_elo (if keep_rating = TRUE) and ranking_elo (Elo ranking computed with round_rank()).

add_elo_ratings() returns a widecr form of cr_data with four rating columns added:

  • rating1Before - Rating of player1 before the game.

  • rating2Before - Rating of player2 before the game.

  • rating1After - Rating of player1 after the game.

  • rating2After - Rating of player2 after the game.

elo() always returns a matrix with two columns containing ratings after the game. Rows represent games, columns - players.

Details

rate_elo() and add_elo_ratings() are wrappers for rate_iterative() and add_iterative_ratings() correspondingly. Rate function is based on Elo algorithm of updating ratings:

  1. Probability of player1 (with rating r1) winning against player2 (with rating r2) is computed based on rating difference and sigmoid function: P = 1 / (1 + 10^( (r2 - r1) / ksi ) ). ksi defines the spread of ratings.

  2. Result of the game from player1 perspective is computed based on rule: S = 1 (if score1 > score2), S = 0.5 (if score1 == score2) and S = 0 (if score1 < score2).

  3. Rating delta is computed: d = K * (S - P). The more the K the more the delta (with other being equal).

  4. New ratings are computed: r1_new = r1 + d, r2_new = r2 - d.

elo() function implements this algorithm. It is vectorized over all its arguments with standard R recycling functionality. Note that not this function is used in rate_elo() and add_elo_ratings() because of its not appropriate output format, but rather its non-vectorized reimplementation is.

Ratings are computed based only on games between players of interest (see Players) and NA values.

Players

comperank offers a possibility to handle certain set of players. It is done by having player column (in longcr format) as factor with levels specifying all players of interest. In case of factor the result is returned only for players from its levels. Otherwise - for all present players.

References

Wikipedia page for Elo rating system.

Examples

# Elo ratings rate_elo(ncaa2005)
#> # A tibble: 5 x 2 #> player rating_elo #> <chr> <dbl> #> 1 Duke -56.2 #> 2 Miami 57.9 #> 3 UNC -1.26 #> 4 UVA -29.2 #> 5 VT 28.8
rank_elo(ncaa2005)
#> # A tibble: 5 x 2 #> player ranking_elo #> <chr> <dbl> #> 1 Duke 5 #> 2 Miami 1 #> 3 UNC 3 #> 4 UVA 4 #> 5 VT 2
rank_elo(ncaa2005, keep_rating = TRUE)
#> # A tibble: 5 x 3 #> player rating_elo ranking_elo #> <chr> <dbl> <dbl> #> 1 Duke -56.2 5 #> 2 Miami 57.9 1 #> 3 UNC -1.26 3 #> 4 UVA -29.2 4 #> 5 VT 28.8 2
add_elo_ratings(ncaa2005, initial_ratings = 100)
#> # A widecr object: #> # A tibble: 10 x 9 #> game player1 score1 player2 score2 rating1Before rating2Before rating1After #> <int> <chr> <int> <chr> <int> <dbl> <dbl> <dbl> #> 1 1 Duke 7 Miami 52 100 100 85 #> 2 2 Duke 21 UNC 24 85 100 70.6 #> 3 3 Duke 7 UVA 38 70.6 100 56.9 #> 4 4 Duke 0 VT 45 56.9 100 43.8 #> 5 5 Miami 34 UNC 16 115 114. 130. #> 6 6 Miami 25 UVA 17 130. 114. 144. #> 7 7 Miami 27 VT 7 144. 113. 158. #> 8 8 UNC 7 UVA 5 99.4 99.4 114. #> 9 9 UNC 3 VT 30 114. 99.5 98.7 #> 10 10 UVA 14 VT 52 84.4 115. 70.8 #> # … with 1 more variable: rating2After <dbl>
# Elo function elo((0:12)*100, 1, 0, 0)
#> [,1] [,2] #> [1,] 15.0000 -15.00000000 #> [2,] 110.7981 -10.79805001 #> [3,] 207.2076 -7.20759220 #> [4,] 304.5294 -4.52938672 #> [5,] 402.7273 -2.72727273 #> [6,] 501.5972 -1.59720646 #> [7,] 600.9196 -0.91960290 #> [8,] 700.5242 -0.52416274 #> [9,] 800.2970 -0.29702970 #> [10,] 900.1678 -0.16775902 #> [11,] 1000.0946 -0.09456928 #> [12,] 1100.0533 -0.05325368 #> [13,] 1200.0300 -0.02997003
elo((0:12)*100, 1, 0, 0, K = 10)
#> [,1] [,2] #> [1,] 5.0000 -5.00000000 #> [2,] 103.5994 -3.59935000 #> [3,] 202.4025 -2.40253073 #> [4,] 301.5098 -1.50979557 #> [5,] 400.9091 -0.90909091 #> [6,] 500.5324 -0.53240215 #> [7,] 600.3065 -0.30653430 #> [8,] 700.1747 -0.17472091 #> [9,] 800.0990 -0.09900990 #> [10,] 900.0559 -0.05591967 #> [11,] 1000.0315 -0.03152309 #> [12,] 1100.0178 -0.01775123 #> [13,] 1200.0100 -0.00999001
elo((0:12)*10, 1, 0, 0, ksi = 40)
#> [,1] [,2] #> [1,] 15.00000 -15.00000000 #> [2,] 20.79805 -10.79805001 #> [3,] 27.20759 -7.20759220 #> [4,] 34.52939 -4.52938672 #> [5,] 42.72727 -2.72727273 #> [6,] 51.59721 -1.59720646 #> [7,] 60.91960 -0.91960290 #> [8,] 70.52416 -0.52416274 #> [9,] 80.29703 -0.29702970 #> [10,] 90.16776 -0.16775902 #> [11,] 100.09457 -0.09456928 #> [12,] 110.05325 -0.05325368 #> [13,] 120.02997 -0.02997003