README.md
In isaacfab/FantasySportsDS:
Fantasy Sports Data Science R tool kit
Isaac J. Faber
November, 2017
This package provides a set of functions to create optimal line-ups for use in fantasy sports leagues. The purpose of this narrative is to show how to use the FantasySportsDS package. First, load the packages.
library(devtools)
library(stringdist)
library(ICSNP)
install_github('isaacfab/FantasySportsDS')
library(FantasySportsDS)
For starters, you will need to read in the data from your league, Draft Kings - NFL is used as the first application. For now the package is limited to functions for the Draft Kings - NFL and the 2016 season. A '.csv' file is provided by the Draft Kings site when you build a line up. We want to create a descriptive name for each player in order to join each row with other information later. Here we also create a variable (binary only 1 or 0) that if set to 1 requires that player to be in a given lineup.
data(DKSalariesExample)
DKSalariesExample$DescriptiveName<-paste(DKSalariesExample$Name,DKSalariesExample$teamAbbrev,DKSalariesExample$Position,sep=" ")
DKSalariesExample$Require<-0
print(head(DKSalariesExample))
## Position Name Salary GameInfo AvgPointsPerGame
## 1 WR Julio Jones 9300 Phi@Atl 07:10PM ET 21.293
## 2 WR Odell Beckham Jr. 9200 NYG@Dal 08:30PM ET 26.417
## 3 WR Demaryius Thomas 9100 Bal@Den 04:25PM ET 22.812
## 4 WR Antonio Brown 8900 Pit@NE 08:30PM ET 25.682
## 5 WR Dez Bryant 8700 NYG@Dal 08:30PM ET 19.033
## 6 QB Aaron Rodgers 8600 GB@Chi 01:00PM ET 23.428
## teamAbbrev DescriptiveName Require
## 1 Atl Julio Jones Atl WR 0
## 2 NYG Odell Beckham Jr. NYG WR 0
## 3 Den Demaryius Thomas Den WR 0
## 4 Pit Antonio Brown Pit WR 0
## 5 Dal Dez Bryant Dal WR 0
## 6 GB Aaron Rodgers GB QB 0
Several things should be observed from the raw data. First, each player has a 'salary' and a line up must be selected such that the total salary does not exceed a given cap (default is 50,000). In addition, a column has the average number of points for a given player. This number is the mean for this season. Draft Kings will include players in the line up selection that are either injured or suspended. You, obviously, want to avoid selecting these players. On the actual website their names have a flag next to them but for analysis or automation we have to remove them ourselves. The first function the FantasySportsDS package provides is an automated method for retrieving current injuries or suspensions (from Yahoo). The first function returns players that are currently ineligible to play. Keep in mind that these functions scrape current data!
data.inel<-FantasySportsDS::injury()
print(data.inel[1:3,])
## Name
## 1 Drew Stanton
## 2 Tyvon Branch
## 3 Carson Palmer
## Description
## 1 kneeESPN's Chris Mortensen reports QB Drew Stanton sprained his knee in Week 10 against the Seahawks.Week to weekNov 11
## 2 kneeCardinals SS Tyvon Branch suffered a torn ACL in Week 10 against the Seahawks and is done for the season.Out for seasonNov 10
## 3 armCarson Palmer (arm, IR) says there's a chance he could return in Week 16.Out indefinitelyNov 11
## POS Status Date Injury Returns DescriptiveName
## 1 QB Sidelined Nov 11 knee Week to week Drew Stanton QB
## 2 DB I.L. Nov 10 knee Out for season Tyvon Branch DB
## 3 QB I.L. Oct 26 arm Out indefinitely Carson Palmer QB
We want to join the injury information to the draft kings DKSalariesExample.
DKSalariesExample$injury<-"None"
DKSalariesExample$gdc<-"None"
for(i in 1:length(DKSalariesExample[,1])){
j<-stringdist::amatch(DKSalariesExample$DescriptiveName[i],data.inel$DescriptiveName,maxDist=5)
if(!is.na(j)){
DKSalariesExample$injury[i]<-data.inel[["Injury"]][j]
}
}
print(DKSalariesExample[1:5,])
## Position Name Salary GameInfo AvgPointsPerGame
## 1 WR Julio Jones 9300 Phi@Atl 07:10PM ET 21.293
## 2 WR Odell Beckham Jr. 9200 NYG@Dal 08:30PM ET 26.417
## 3 WR Demaryius Thomas 9100 Bal@Den 04:25PM ET 22.812
## 4 WR Antonio Brown 8900 Pit@NE 08:30PM ET 25.682
## 5 WR Dez Bryant 8700 NYG@Dal 08:30PM ET 19.033
## teamAbbrev DescriptiveName Require injury gdc
## 1 Atl Julio Jones Atl WR 0 None None
## 2 NYG Odell Beckham Jr. NYG WR 0 None None
## 3 Den Demaryius Thomas Den WR 0 None None
## 4 Pit Antonio Brown Pit WR 0 None None
## 5 Dal Dez Bryant Dal WR 0 None None
Now you have to make some decisions on what players you want to exclude. Obviously, you want to rule out anyone who is 'Out' but there is also a category of 'Questionable.' This category often includes high profile players that may need a case-by-case decision. For this example, we will only keep players that are either not injured or are probable. You can subset however you like.
DKSalariesExample<-DKSalariesExample[DKSalariesExample$gdc=="None"|DKSalariesExample$gdc=="P",]
table(DKSalariesExample$injury)
##
## - achilles ankle arm back
## 9 1 8 3 2
## biceps calf clavicle concussion elbow
## 1 1 1 7 1
## foot groin hamstring hip illness
## 2 3 7 1 2
## knee knee surgery leg neck None
## 13 9 1 3 862
## shoulder toe undisclosed wrist
## 6 1 1 3
Now that you have removed any players that will likely not be on the field, it is time to start thinking about how to forecast player performance. The following function scrapes fantasy point predictions from 8 fantasy football websites and gives two types of averages. The only arguments that this function accepts is an integer value for the week of the season and the year. Like the injuries, this function returns current values.
forecast<-FantasySportsDS::CombinedForecasts(1,2015)
print(forecast[1:3,])
## DescriptiveName FFToday_f CBS_f Yahoo_f NFL_f FFPC_f NFFC_f
## 2 Aaron Rodgers GB QB 30.0 32.7 26.7 32.7 30.0 36.0
## 3 Tony Romo DAL QB 27.3 30.3 24.3 30.3 27.3 33.3
## 4 Tom Brady NE QB 27.0 28.0 23.0 28.0 26.0 32.0
## Average_Robust Mean SD
## 2 31.35 31.35000 3.179151
## 3 28.80 28.80000 3.146427
## 4 27.50 27.33333 2.943920
Need to combine these forecasts with the data.
n<-colnames(forecast)
n<-n[2:length(n)]
for(i in 1:length(n)){
DKSalariesExample[[n[i]]]<-0
}
#combine the forecasts and the Draft Kings file
for(i in 1:length(DKSalariesExample[,1])){
j<-stringdist::amatch(DKSalariesExample$DescriptiveName[i],forecast$DescriptiveName,maxDist=5)
if(!is.na(j)){
for(k in 1:length(n)){
DKSalariesExample[[n[k]]][i]<-forecast[[k+1]][j]
}
}
}
The quality of the lineup that is played is dependent on how good the forecasts are. There is room here for you to complete your own forecasts from any number of sources such as regression, machine learning, other expert predictions, ownership rates and so on. However, even if you have perfect forecasts you would still have the problem of knowing which combination of players results in the greatest amount of fantasy points (the possible combinations of teams is enormous).
The following function preforms a linear program and returns the optimal lineup from a given set of forecasts. In other words if your forecast are perfect this function will give you the team with the highest score. There are three functional arguments; our data, the forecast column name and the salary cap (defaults to 50,000). Here are three optimal lineups using different sources of forecasting and constraints.
#Optimal Lineup with Average Points Per Game forecast
opt<-FantasySportsDS::DKoptLineUpNFL(DKSalariesExample,"AvgPointsPerGame")
print(opt[,1:4])
## Position Name Salary GameInfo
## 95 QB Mark Sanchez 5400 Phi@Atl 07:10PM ET
## 51 RB DeMarco Murray 6700 Phi@Atl 07:10PM ET
## 349 RB Ahmad Bradshaw 3000 Ind@Buf 01:00PM ET
## 2 WR Odell Beckham Jr. 9200 NYG@Dal 08:30PM ET
## 4 WR Antonio Brown 8900 Pit@NE 08:30PM ET
## 243 WR Martavis Bryant 3900 Pit@NE 08:30PM ET
## 801 TE Dennis Pitta 2500 Bal@Den 04:25PM ET
## 25 FLEX Le'Veon Bell 7600 Pit@NE 08:30PM ET
## 783 DST Eagles 2700 Phi@Atl 07:10PM ET
#Optimal Lineup with FFToday forecast and require that Russell Wilson is in the lineup
DKSalariesExample$Require[DKSalariesExample$DescriptiveName=="Russell Wilson Sea QB"]<-1
fft_opt<-FantasySportsDS::DKoptLineUpNFL(DKSalariesExample,"FFToday_f")
print(fft_opt[,1:4])
## Position Name Salary GameInfo
## 17 QB Russell Wilson 7800 Sea@StL 01:00PM ET
## 212 RB Doug Martin 4500 Ten@TB 04:25PM ET
## 232 RB Chris Ivory 4100 Cle@NYJ 01:00PM ET
## 13 WR Randall Cobb 8000 GB@Chi 01:00PM ET
## 218 WR Davante Adams 4400 GB@Chi 01:00PM ET
## 255 WR Stevie Johnson 3700 Det@SD 04:05PM ET
## 43 TE Rob Gronkowski 7000 Pit@NE 08:30PM ET
## 20 FLEX Adrian Peterson 7700 Min@SF 10:20PM ET
## 948 DST Raiders 2300 Cin@Oak 04:25PM ET
#Optimal Lineup with Average forecast but lower salary cap
avg_opt<-FantasySportsDS::DKoptLineUpNFL(DKSalariesExample,"Average_Robust",45000)
print(avg_opt[,1:4])
## Position Name Salary GameInfo
## 17 QB Russell Wilson 7800 Sea@StL 01:00PM ET
## 212 RB Doug Martin 4500 Ten@TB 04:25PM ET
## 232 RB Chris Ivory 4100 Cle@NYJ 01:00PM ET
## 218 WR Davante Adams 4400 GB@Chi 01:00PM ET
## 255 WR Stevie Johnson 3700 Det@SD 04:05PM ET
## 449 WR Jaron Brown 3000 NO@Ari 04:05PM ET
## 43 TE Rob Gronkowski 7000 Pit@NE 08:30PM ET
## 13 FLEX Randall Cobb 8000 GB@Chi 01:00PM ET
## 948 DST Raiders 2300 Cin@Oak 04:25PM ET
isaacfab/FantasySportsDS documentation built on May 28, 2019, 9:56 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Fantasy Sports Data Science R tool kit
Isaac J. Faber November, 2017
This package provides a set of functions to create optimal line-ups for use in fantasy sports leagues. The purpose of this narrative is to show how to use the FantasySportsDS package. First, load the packages.
library(devtools)
library(stringdist)
library(ICSNP)
install_github('isaacfab/FantasySportsDS')
library(FantasySportsDS)
For starters, you will need to read in the data from your league, Draft Kings - NFL is used as the first application. For now the package is limited to functions for the Draft Kings - NFL and the 2016 season. A '.csv' file is provided by the Draft Kings site when you build a line up. We want to create a descriptive name for each player in order to join each row with other information later. Here we also create a variable (binary only 1 or 0) that if set to 1 requires that player to be in a given lineup.
data(DKSalariesExample)
DKSalariesExample$DescriptiveName<-paste(DKSalariesExample$Name,DKSalariesExample$teamAbbrev,DKSalariesExample$Position,sep=" ")
DKSalariesExample$Require<-0
print(head(DKSalariesExample))
## Position Name Salary GameInfo AvgPointsPerGame
## 1 WR Julio Jones 9300 Phi@Atl 07:10PM ET 21.293
## 2 WR Odell Beckham Jr. 9200 NYG@Dal 08:30PM ET 26.417
## 3 WR Demaryius Thomas 9100 Bal@Den 04:25PM ET 22.812
## 4 WR Antonio Brown 8900 Pit@NE 08:30PM ET 25.682
## 5 WR Dez Bryant 8700 NYG@Dal 08:30PM ET 19.033
## 6 QB Aaron Rodgers 8600 GB@Chi 01:00PM ET 23.428
## teamAbbrev DescriptiveName Require
## 1 Atl Julio Jones Atl WR 0
## 2 NYG Odell Beckham Jr. NYG WR 0
## 3 Den Demaryius Thomas Den WR 0
## 4 Pit Antonio Brown Pit WR 0
## 5 Dal Dez Bryant Dal WR 0
## 6 GB Aaron Rodgers GB QB 0
Several things should be observed from the raw data. First, each player has a 'salary' and a line up must be selected such that the total salary does not exceed a given cap (default is 50,000). In addition, a column has the average number of points for a given player. This number is the mean for this season. Draft Kings will include players in the line up selection that are either injured or suspended. You, obviously, want to avoid selecting these players. On the actual website their names have a flag next to them but for analysis or automation we have to remove them ourselves. The first function the FantasySportsDS package provides is an automated method for retrieving current injuries or suspensions (from Yahoo). The first function returns players that are currently ineligible to play. Keep in mind that these functions scrape current data!
data.inel<-FantasySportsDS::injury()
print(data.inel[1:3,])
## Name
## 1 Drew Stanton
## 2 Tyvon Branch
## 3 Carson Palmer
## Description
## 1 kneeESPN's Chris Mortensen reports QB Drew Stanton sprained his knee in Week 10 against the Seahawks.Week to weekNov 11
## 2 kneeCardinals SS Tyvon Branch suffered a torn ACL in Week 10 against the Seahawks and is done for the season.Out for seasonNov 10
## 3 armCarson Palmer (arm, IR) says there's a chance he could return in Week 16.Out indefinitelyNov 11
## POS Status Date Injury Returns DescriptiveName
## 1 QB Sidelined Nov 11 knee Week to week Drew Stanton QB
## 2 DB I.L. Nov 10 knee Out for season Tyvon Branch DB
## 3 QB I.L. Oct 26 arm Out indefinitely Carson Palmer QB
We want to join the injury information to the draft kings DKSalariesExample.
DKSalariesExample$injury<-"None"
DKSalariesExample$gdc<-"None"
for(i in 1:length(DKSalariesExample[,1])){
j<-stringdist::amatch(DKSalariesExample$DescriptiveName[i],data.inel$DescriptiveName,maxDist=5)
if(!is.na(j)){
DKSalariesExample$injury[i]<-data.inel[["Injury"]][j]
}
}
print(DKSalariesExample[1:5,])
## Position Name Salary GameInfo AvgPointsPerGame
## 1 WR Julio Jones 9300 Phi@Atl 07:10PM ET 21.293
## 2 WR Odell Beckham Jr. 9200 NYG@Dal 08:30PM ET 26.417
## 3 WR Demaryius Thomas 9100 Bal@Den 04:25PM ET 22.812
## 4 WR Antonio Brown 8900 Pit@NE 08:30PM ET 25.682
## 5 WR Dez Bryant 8700 NYG@Dal 08:30PM ET 19.033
## teamAbbrev DescriptiveName Require injury gdc
## 1 Atl Julio Jones Atl WR 0 None None
## 2 NYG Odell Beckham Jr. NYG WR 0 None None
## 3 Den Demaryius Thomas Den WR 0 None None
## 4 Pit Antonio Brown Pit WR 0 None None
## 5 Dal Dez Bryant Dal WR 0 None None
Now you have to make some decisions on what players you want to exclude. Obviously, you want to rule out anyone who is 'Out' but there is also a category of 'Questionable.' This category often includes high profile players that may need a case-by-case decision. For this example, we will only keep players that are either not injured or are probable. You can subset however you like.
DKSalariesExample<-DKSalariesExample[DKSalariesExample$gdc=="None"|DKSalariesExample$gdc=="P",]
table(DKSalariesExample$injury)
##
## - achilles ankle arm back
## 9 1 8 3 2
## biceps calf clavicle concussion elbow
## 1 1 1 7 1
## foot groin hamstring hip illness
## 2 3 7 1 2
## knee knee surgery leg neck None
## 13 9 1 3 862
## shoulder toe undisclosed wrist
## 6 1 1 3
Now that you have removed any players that will likely not be on the field, it is time to start thinking about how to forecast player performance. The following function scrapes fantasy point predictions from 8 fantasy football websites and gives two types of averages. The only arguments that this function accepts is an integer value for the week of the season and the year. Like the injuries, this function returns current values.
forecast<-FantasySportsDS::CombinedForecasts(1,2015)
print(forecast[1:3,])
## DescriptiveName FFToday_f CBS_f Yahoo_f NFL_f FFPC_f NFFC_f
## 2 Aaron Rodgers GB QB 30.0 32.7 26.7 32.7 30.0 36.0
## 3 Tony Romo DAL QB 27.3 30.3 24.3 30.3 27.3 33.3
## 4 Tom Brady NE QB 27.0 28.0 23.0 28.0 26.0 32.0
## Average_Robust Mean SD
## 2 31.35 31.35000 3.179151
## 3 28.80 28.80000 3.146427
## 4 27.50 27.33333 2.943920
Need to combine these forecasts with the data.
n<-colnames(forecast)
n<-n[2:length(n)]
for(i in 1:length(n)){
DKSalariesExample[[n[i]]]<-0
}
#combine the forecasts and the Draft Kings file
for(i in 1:length(DKSalariesExample[,1])){
j<-stringdist::amatch(DKSalariesExample$DescriptiveName[i],forecast$DescriptiveName,maxDist=5)
if(!is.na(j)){
for(k in 1:length(n)){
DKSalariesExample[[n[k]]][i]<-forecast[[k+1]][j]
}
}
}
The quality of the lineup that is played is dependent on how good the forecasts are. There is room here for you to complete your own forecasts from any number of sources such as regression, machine learning, other expert predictions, ownership rates and so on. However, even if you have perfect forecasts you would still have the problem of knowing which combination of players results in the greatest amount of fantasy points (the possible combinations of teams is enormous).
The following function preforms a linear program and returns the optimal lineup from a given set of forecasts. In other words if your forecast are perfect this function will give you the team with the highest score. There are three functional arguments; our data, the forecast column name and the salary cap (defaults to 50,000). Here are three optimal lineups using different sources of forecasting and constraints.
#Optimal Lineup with Average Points Per Game forecast
opt<-FantasySportsDS::DKoptLineUpNFL(DKSalariesExample,"AvgPointsPerGame")
print(opt[,1:4])
## Position Name Salary GameInfo
## 95 QB Mark Sanchez 5400 Phi@Atl 07:10PM ET
## 51 RB DeMarco Murray 6700 Phi@Atl 07:10PM ET
## 349 RB Ahmad Bradshaw 3000 Ind@Buf 01:00PM ET
## 2 WR Odell Beckham Jr. 9200 NYG@Dal 08:30PM ET
## 4 WR Antonio Brown 8900 Pit@NE 08:30PM ET
## 243 WR Martavis Bryant 3900 Pit@NE 08:30PM ET
## 801 TE Dennis Pitta 2500 Bal@Den 04:25PM ET
## 25 FLEX Le'Veon Bell 7600 Pit@NE 08:30PM ET
## 783 DST Eagles 2700 Phi@Atl 07:10PM ET
#Optimal Lineup with FFToday forecast and require that Russell Wilson is in the lineup
DKSalariesExample$Require[DKSalariesExample$DescriptiveName=="Russell Wilson Sea QB"]<-1
fft_opt<-FantasySportsDS::DKoptLineUpNFL(DKSalariesExample,"FFToday_f")
print(fft_opt[,1:4])
## Position Name Salary GameInfo
## 17 QB Russell Wilson 7800 Sea@StL 01:00PM ET
## 212 RB Doug Martin 4500 Ten@TB 04:25PM ET
## 232 RB Chris Ivory 4100 Cle@NYJ 01:00PM ET
## 13 WR Randall Cobb 8000 GB@Chi 01:00PM ET
## 218 WR Davante Adams 4400 GB@Chi 01:00PM ET
## 255 WR Stevie Johnson 3700 Det@SD 04:05PM ET
## 43 TE Rob Gronkowski 7000 Pit@NE 08:30PM ET
## 20 FLEX Adrian Peterson 7700 Min@SF 10:20PM ET
## 948 DST Raiders 2300 Cin@Oak 04:25PM ET
#Optimal Lineup with Average forecast but lower salary cap
avg_opt<-FantasySportsDS::DKoptLineUpNFL(DKSalariesExample,"Average_Robust",45000)
print(avg_opt[,1:4])
## Position Name Salary GameInfo
## 17 QB Russell Wilson 7800 Sea@StL 01:00PM ET
## 212 RB Doug Martin 4500 Ten@TB 04:25PM ET
## 232 RB Chris Ivory 4100 Cle@NYJ 01:00PM ET
## 218 WR Davante Adams 4400 GB@Chi 01:00PM ET
## 255 WR Stevie Johnson 3700 Det@SD 04:05PM ET
## 449 WR Jaron Brown 3000 NO@Ari 04:05PM ET
## 43 TE Rob Gronkowski 7000 Pit@NE 08:30PM ET
## 13 FLEX Randall Cobb 8000 GB@Chi 01:00PM ET
## 948 DST Raiders 2300 Cin@Oak 04:25PM ET
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.