| Type: | Package |
| Title: | Evaluating Sport Tournament Predictions |
| Version: | 0.1.1 |
| Date: | 2019-07-01 |
| Maintainer: | Claus Thorn Ekstrøm <ekstrom@sund.ku.dk> |
| Description: | Functions for evaluating tournament predictions, simulating results from individual soccer matches and tournaments. See http://sandsynligvis.dk/2018/08/03/world-cup-prediction-winners/ for more information. |
| License: | GPL-2 | GPL-3 [expanded from: GPL (≥ 2)] |
| Depends: | R (≥ 3.1.0) |
| Imports: | Rcpp (≥ 1.0.0) |
| LinkingTo: | Rcpp |
| LazyData: | true |
| RoxygenNote: | 6.1.1 |
| Encoding: | UTF-8 |
| URL: | https://github.com/ekstroem/socceR |
| BugReports: | https://github.com/ekstroem/socceR/issues |
| NeedsCompilation: | yes |
| Packaged: | 2019-07-02 09:09:01 UTC; cld189 |
| Author: | Claus Thorn Ekstrøm [aut, cre] |
| Repository: | CRAN |
| Date/Publication: | 2019-07-03 11:50:03 UTC |
Evaluating sport tournament predictions
Description
Functions for evaluating sport tournament predictions, the tournament rank probability score, and working with models for prediction sport matches.
Author(s)
Claus Ekstrom <ekstrom@sund.ku.dk>
Create a matrix to collapse tournament predictions to ranks
Description
Creates a matrix to collapse the rows of a tournamewnt prediction matrix
Usage
collapse_prediction(ranks = c(1, 2, 3, 4, 8, 16, 32))
Arguments
ranks |
An integer vector of R ordered elements giving the cut offs of the ranks to create |
Details
Returns a vector of numeric values. Elements in the input factor that cannot be converted to numeric will produce NA.
Value
Returns a numeric matrix with R rows and T columns that can be multiplied on a square prediction matrix to obtain the collapsed predictions
Author(s)
Claus Ekstrom ekstrom@sund.ku.dk
Examples
m2 <- matrix(c(.5, .5, 0, 0, .5, .5, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1), 4)
# Collapse into ranks 1, 2, and 3+4
collapse <- collapse_prediction(c(1, 2, 4))
collapsed_prediction <- collapse %*% m2
collapsed_prediction
FIFA 2018 prediction matrices
Description
A list containing five predictions for the FIFA 2018 World Cup.
Usage
fifa2018
Format
A list with 5 predictions (each a 7 by 32 matrix) containing the predictions probabilities of 1st, 2nd, 3rd, 4th, 5th-8th, 9th-12th, and 17th-32nd place.
- flat
A prediction with equal probability of winning for all teams
- ekstrom1
Ekstrom's prediction (based on the Skellam distribution)
- ekstrom2
Ekstrom's prediction (based on the ELO rankings)
- GLSE1
Prediction of Groll et all
- GLSE2
Updated prediction of Groll et all
Source
http://sandsynligvis.dk/2018/08/03/world-cup-prediction-winners/
FIFA 2018 end results
Description
A named vector sorted in the ranking of the teams in the FIFA 2018 World Cup. The value correspond to the corresponding columns in the prediction matrices of fifa2018
Usage
fifa2018result
Format
A vector of the final rankings
Source
http://sandsynligvis.dk/2018/08/03/world-cup-prediction-winners/
Computes the log loss score for a tournament prediction
Description
Compute the (weighted) rank probability score for a tournament.
Usage
logloss(m, outcome, rankweights = 1L)
Arguments
m |
An R*T prediction matrix where the R rows represent the ordered ranks and each column is a team. Each column should sum to 1, and each row should sum to the number of teams that can attain a given rank. |
outcome |
A vector of length T containing the integers 1 to R giving the ranks that were obtained by each of the T teams |
rankweights |
A vector of length R of rank weights or a single weight which will be reused for all ranks (defaults to 1) |
Value
The rank probability score. Zero means a perfect score.
Author(s)
Claus Ekstrom <ekstrom@sund.ku.dk>
Examples
m1 <- matrix(c(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, .5, .5, 0, 0, .5, .5), 4)
m1 # Prediction where certain on the top ranks
logloss(m1, c(1, 2, 3, 4))
Optimize weights from list of prediction matrices
Description
Computes the optimal weights to obtain the minimal loss function from a list of prediction matrices.
Usage
optimize_weights(predictionlist, outcome, FUN = trps)
Arguments
predictionlist |
A list of R x T prediction matrices where each column sum to 1 and each row sums to |
outcome |
An integer vector listing the |
FUN |
The function used for optimizing the predictions. The default is top use rps for the rank probability score. Another option is logloss for log loss. |
Value
Returns a numeric vector containing an optimal vector of weights that sum to 1 and that minimizes the loss function.
Author(s)
Claus Ekstrom ekstrom@sund.ku.dk
Examples
m1 <- matrix(c(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, .5, .5, 0, 0, .5, .5), 4)
m1 # Prediction where certain on the top ranks
m2 <- matrix(c(.5, .5, 0, 0, .5, .5, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1), 4)
m2 # Prediction where the groups are okay
m3 <- matrix(c(.5, .5, 0, 0, .5, .5, 0, 0, 0, 0, .5, .5, 0, 0, .5, .5), 4)
m3 # Prediction where no clue about anything
m4 <- matrix(rep(1/4, 16), 4)
optimize_weights(list(m1, m2, m3, m4), 1:4)
Computes the rank probability score for a tournament
Description
Compute the (weighted) rank probability score for a tournament.
Usage
trps(m, outcome, rankweights = 1L)
Arguments
m |
An R*T prediction matrix where the R rows represent the ordered ranks and each column is a team. Each column should sum to 1, and each row should sum to the number of teams that can attain a given rank. |
outcome |
A vector of length T containing the integers 1 to R giving the ranks that were obtained by each of the T teams |
rankweights |
A vector of length R of rank weights or a single weight which will be reused for all ranks (defaults to 1) |
Value
The rank probability score. Zero means a perfect score.
Author(s)
Claus Ekstrom <ekstrom@sund.ku.dk>
Examples
m1 <- matrix(c(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, .5, .5, 0, 0, .5, .5), 4)
m1 # Prediction where certain on the top ranks
trps(m1, c(1, 2, 3, 4))