README.md

In jedalong/nhlmodelr:

output: github_document

nhlmodelr: Spatial Modelling of NHL Shot Location Data

Access pre-formatted NHL shot location data and implement spatially explicit models of NHL goal scoring probability. Advanced team, player, and goaltender statistics can be automatically calculated.

Installation

devtools::install_github("jedalong/nhlmodelr")

Quick demo

Load in some data (2011-12 season) and take a look at the structure.

library(nhlmodelr)
data(shots20112012) 
head(shots20112012)
#>   PLAYID PLYR1.ID       PlyrName PlyrPos TeamID TeamStrength ShotType
#> 1    506      246   Patrik Elias      LW     11          701      805
#> 2    505     3343   David Krejci      C       1          701      803
#> 3    508      816 Bryce Salvador      D      11          701        0
#> 4    508     5490     Mark Fayne      D      11          701        0
#> 5    508     5430   Tyler Seguin      C       1          701        0
#> 6    507     5430   Tyler Seguin      C       1          701      802
#>   ShotDistFt GOALIE.ID     GoalieName gTeamID Goalie   x   y   SEASON
#> 1         14      1207     Tim Thomas       1    901 -76   1 20112012
#> 2         12       108 Martin Brodeur      11    901 -78   1 20112012
#> 3         NA        NA           <NA>       1      0 -70  -7 20112012
#> 4         NA        NA           <NA>       1      0 -42  24 20112012
#> 5         NA        NA           <NA>      11      0 -69 -18 20112012
#> 6         39        NA           <NA>      11    901 -51   0 20112012
#>   PreRegPost    GameID
#> 1          2 400046681
#> 2          2 400046681
#> 3          2 400046681
#> 4          2 400046681
#> 5          2 400046681
#> 6          2 400046681
summary(shots20112012)
#>      PLAYID         PLYR1.ID         PlyrName           PlyrPos         
#>  Min.   :505.0   Min.   :      0   Length:142668      Length:142668     
#>  1st Qu.:506.0   1st Qu.:   1576   Class :character   Class :character  
#>  Median :506.0   Median :   3127   Mode  :character   Mode  :character  
#>  Mean   :506.7   Mean   :  53534                                        
#>  3rd Qu.:508.0   3rd Qu.:   3792                                        
#>  Max.   :508.0   Max.   :2953322                                        
#>                                                                         
#>      TeamID       TeamStrength      ShotType       ShotDistFt    
#>  Min.   : 1.00   Min.   :  0.0   Min.   :  0.0   Min.   :  2.00  
#>  1st Qu.: 8.00   1st Qu.:701.0   1st Qu.:  0.0   1st Qu.: 18.00  
#>  Median :15.00   Median :701.0   Median :802.0   Median : 35.00  
#>  Mean   :15.31   Mean   :696.5   Mean   :598.7   Mean   : 35.89  
#>  3rd Qu.:23.00   3rd Qu.:701.0   3rd Qu.:803.0   3rd Qu.: 50.00  
#>  Max.   :30.00   Max.   :703.0   Max.   :807.0   Max.   :189.00  
#>                                                  NA's   :36294   
#>    GOALIE.ID        GoalieName           gTeamID          Goalie     
#>  Min.   :     72   Length:142668      Min.   : 1.00   Min.   :  0.0  
#>  1st Qu.:   1207   Class :character   1st Qu.: 8.00   1st Qu.:  0.0  
#>  Median :   2339   Mode  :character   Median :15.00   Median :901.0  
#>  Mean   :   2459                      Mean   :15.62   Mean   :672.1  
#>  3rd Qu.:   3502                      3rd Qu.:23.00   3rd Qu.:901.0  
#>  Max.   :2590376                      Max.   :30.00   Max.   :905.0  
#>  NA's   :66292                                                       
#>        x                y                SEASON           PreRegPost   
#>  Min.   :-99.00   Min.   :-42.0000   Min.   :20112012   Min.   :2.000  
#>  1st Qu.:-76.00   1st Qu.:-12.0000   1st Qu.:20112012   1st Qu.:2.000  
#>  Median :-64.00   Median :  1.0000   Median :20112012   Median :2.000  
#>  Mean   :-61.24   Mean   :  0.4313   Mean   :20112045   Mean   :2.068  
#>  3rd Qu.:-49.00   3rd Qu.: 13.0000   3rd Qu.:20112012   3rd Qu.:2.000  
#>  Max.   :  0.00   Max.   : 42.0000   Max.   :20122013   Max.   :3.000  
#>                                                                        
#>      GameID         
#>  Min.   :400046681  
#>  1st Qu.:400047008  
#>  Median :400047343  
#>  Mean   :400061787  
#>  3rd Qu.:400047671  
#>  Max.   :400265531  
#> 

Next, format the shots for the spatial model by spatially pooling all the shots to the 5525 unique locations in the offensive zone.

modshots <- shots2model(shots20112012)
head(modshots)
#>     x   y xi ni
#> 1 -89 -42  0  0
#> 2 -88 -42  0  0
#> 3 -87 -42  0  0
#> 4 -86 -42  0  0
#> 5 -85 -42  0  0
#> 6 -84 -42  0  0

Set up parameters for the spatial model. First, we need to set the number of Gibbs samples for the MCMC chain and the burn-in. We also need to set-up initial values for each location (here we use random values between 0 and 1). The parameter p.max is the tuning parameter, which is the maximum probability of scoring a goal (in the centre of the goal). Finally, the output filename for storing output *.RData file needs to be specified (the model produces a single *.RData file with 8 objects, see ?NHLmodelr).

ngibbs <- 100
nburnin <- 50
p.max <- 0.29
inits <- runif(dim(modshots)[1],0,1)
outfile <- 'C:/Workspace/NHLModelOutput.RData'

#Model Run - Takes a couple of minutes
df <- NHLmodel(modshots,ngibbs,nburnin,p.max,inits,outfile)
head(df)
#>   x y oldx oldy newx newy      rad    theta index n g p_postmean
#> 1 1 1  -89  -42    0  -42 42.00000 1.570796     1 0 0 0.01132139
#> 2 2 1  -88  -42    1  -42 42.01190 1.546991     2 0 0 0.01631660
#> 3 3 1  -87  -42    2  -42 42.04759 1.523213     3 0 0 0.01599074
#> 4 4 1  -86  -42    3  -42 42.10701 1.499489     4 0 0 0.01696423
#> 5 5 1  -85  -42    4  -42 42.19005 1.475845     5 0 0 0.01598282
#> 6 6 1  -84  -42    5  -42 42.29657 1.452306     6 0 0 0.02143530
#>      p_postvar
#> 1 1.133051e-04
#> 2 1.449830e-04
#> 3 9.757342e-05
#> 4 1.591031e-04
#> 5 1.384406e-04
#> 6 2.437138e-04

Their are a number of plotting functions available. We can plot the raw shots data, only the goals, the posterior mean, and the posterior variance. Note: The posterior mean and variance here are unrealistic as they are based on only 100 Gibbs samples from the MCMC run.

NHLmodelplot(df,'shots')

NHLmodelplot(df,'goals')

NHLmodelplot(df,'postmean')

NHLmodelplot(df,'postvar')

We can also compute advanced team, player, and goaltender spatially adjusted scoring (and save percentage) statistics.

#==== Not Run ====
# teamdb <- teams(df,shots20112012)
# playerdb <- players(df,shots20112012)
# goaliedb <- goalies(df,shots20112012)
#====

--- End ---

jedalong/nhlmodelr documentation built on May 25, 2019, 6:23 p.m.