Winshares.Rmd

---
title: "Final Project"
author: "Ray"
date: "March, 31, 2020"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)

```

Section 1-Predicting Current Season Win Shares

```{r}
#Importing Data
setwd("D:/02Grad School/R-Dir/data")
NBA= read.csv("Seasons_stats_complete.csv",stringsAsFactors=TRUE)
str(NBA)
```

```{r}
#Exploring the data, numerical and categorical variables 

library(dplyr)
library(ggplot2)
library(tidyverse)

#Selecting correlated numerical variables
num1<-dplyr::select_if(NBA, is.numeric)
num1
WinShares=NBA$WS
corr1<-cor(WinShares, num1)
corr1
correlated1= which(corr1>0)
correlated1
colnames(correlated1)
summary(correlated1)

#Scatterplot of Correlated Numerical Variables
ggplot(NBA,aes(x=Year, y=WS)) + geom_point()
ggplot(NBA,aes(x=Age, y=WS)) + geom_point()
ggplot(NBA,aes(x=BLK, y=WS)) + geom_point()
ggplot(NBA,aes(x=G, y=WS)) + geom_point()
ggplot(NBA,aes(x=PER, y=WS)) + geom_point()
ggplot(NBA,aes(x=TS., y=WS)) + geom_point()
ggplot(NBA,aes(x=X3PAr, y=WS)) + geom_point()
ggplot(NBA,aes(x=FTr, y=WS)) + geom_point()
ggplot(NBA,aes(x=BLK., y=WS)) + geom_point()
ggplot(NBA,aes(x=ORB., y=WS)) + geom_point()
ggplot(NBA,aes(x=DRB., y=WS)) + geom_point()
ggplot(NBA,aes(x=TRB., y=WS)) + geom_point()
ggplot(NBA,aes(x=STL., y=WS)) + geom_point()
ggplot(NBA,aes(x=TOV., y=WS)) + geom_point()
ggplot(NBA,aes(x=USG., y=WS)) + geom_point()
ggplot(NBA,aes(x=OWS, y=WS)) + geom_point()
ggplot(NBA,aes(x=VORP, y=WS)) + geom_point()
ggplot(NBA,aes(x=DWS, y=WS)) + geom_point()
ggplot(NBA,aes(x=WS.48, y=WS)) + geom_point()
ggplot(NBA,aes(x=OBPM, y=WS)) + geom_point()
ggplot(NBA,aes(x=DBPM, y=WS)) + geom_point()
ggplot(NBA,aes(x=BPM, y=WS)) + geom_point()
ggplot(NBA,aes(x=X3P, y=WS)) + geom_point()
ggplot(NBA,aes(x=X3P., y=WS)) + geom_point()
ggplot(NBA,aes(x=X3PA, y=WS)) + geom_point()
ggplot(NBA,aes(x=X2PA, y=WS)) + geom_point()
ggplot(NBA,aes(x=X2P., y=WS)) + geom_point()
ggplot(NBA,aes(x=X2P, y=WS)) + geom_point()
ggplot(NBA,aes(x=FT, y=WS)) + geom_point()
ggplot(NBA,aes(x=FTA, y=WS)) + geom_point()
ggplot(NBA,aes(x=FT., y=WS)) + geom_point()
ggplot(NBA,aes(x=ORB, y=WS)) + geom_point()
ggplot(NBA,aes(x=DRB, y=WS)) + geom_point()
ggplot(NBA,aes(x=TRB, y=WS)) + geom_point()
ggplot(NBA,aes(x=AST, y=WS)) + geom_point()
ggplot(NBA,aes(x=STL, y=WS)) + geom_point()
ggplot(NBA,aes(x=FG., y=WS)) + geom_point()
ggplot(NBA,aes(x=FG, y=WS)) + geom_point()
ggplot(NBA,aes(x=FGA, y=WS)) + geom_point()
ggplot(NBA,aes(x=eFG., y=WS)) + geom_point()

#Selecting categorical variables
categ= NBA %>% select_if(negate(is.numeric))
str(categ)

#Bar plots of Categorical data
ggplot(NBA, aes(x=Pos,y=WinShares)) + geom_bar(stat="identity") + theme(axis.text.x = element_text(angle = 90, hjust = 1))
ggplot(NBA, aes(x=year,y=WinShares)) + geom_bar(stat="identity") + theme(axis.text.x = element_text(angle = 90, hjust = 1))
ggplot(NBA, aes(x=Player,y=WinShares)) + geom_bar(stat="identity") + theme(axis.text.x = element_text(angle = 90, hjust = 1))

#Boxplots of Categorical data
plot(NBA$Pos ~ WinShares)
plot(NBA$Tm ~ WinShares)
plot(NBA$Player ~ WinShares )

#Histogram: This shows that Win Shares are very right skewed. Many players do not contribute at all due to lack of playing time, injuries, etc.
NBA1<-select(NBA, -contains("WinShares"))
ggplot(NBA1, aes(x=WinShares)) + geom_histogram()

#Summary of the WinShares variable
summary(WinShares)

#Summary of minutes played by NBA players
summary(NBA$MP)
```

```{r}
#Selecting superset of variables from the nba set 
library(dplyr)
NBA=NBA %>% 
  arrange(Year, Player) %>% 
  group_by(Player) %>% 
  arrange(Year) %>%
  #Not including Defensive or Offensive Win Shares because they add up Win Shares directly.
  select(Tm, Pos, Player, Year, Age, MP, G, PER, TS., X3PAr, FTr, ORB., TRB., DRB., STL., BLK., TOV., USG., OBPM, BPM, DBPM, VORP, FG, FGA, FG., X3P, X3PA, X3P., X2P, X2PA, X2P., eFG., FT, FTA, FT., ORB, DRB, TRB, AST, STL, BLK, WS) %>% 
filter(Year >= 1977)%>%  
#294 minutes per game is the 1st quantile of minutes played per season. Many players do not play a single minute in the NBA but are on a team.
filter(MP >= 294)

```
```{r}
#Checking Missing Cases
missing<-sum(!complete.cases(NBA))
missing

cm=colMeans(is.na(NBA))
cm
counts=which(cm>0.0)
counts
length(counts)
cm[counts]


WinShares = NBA$WS

#Scaling, Taking out Categorical variables, Year, and the Winshare dependent variables
NBA1<- as.data.frame(scale(NBA[, -c(1, 2, 3, 42)]))

#Binding the variables together
NBA = cbind(NBA1, WinShares)

#Checking for missing cases
missing<-sum(!complete.cases(WinShares))
missing


```


```{r}
#Splitting the data
set.seed(1)
library(caret)

indexTrain <- createDataPartition(NBA$WinShares, p = 0.9,list = FALSE)
NBA_train <- NBA[indexTrain,]
NBA_test <- NBA[-indexTrain,]

indexTrainValidation <- createDataPartition(y = NBA_train$WinShares, p = 0.9,list = FALSE)
NBA_val <- NBA_train[-indexTrainValidation,]
```

```{r}
#Linear Regression

set.seed(1)
library(caret)
train.control = trainControl(method = "cv", number = 10)
linearRegression <- train(WinShares ~., data = NBA_train, method = "lm", trControl = train.control)
print(linearRegression)
summary(linearRegression)

#Checking the RMSE of the WinShares of the current year
predictionsLR= predict(linearRegression, NBA_test)
RMSE(predictionsLR, NBA_test$WinShares)
R2(predictionsLR, NBA_test$WinShares)

#Comparing to the Standard Deviation of the WinShares Column
sd(NBA_train$WinShares)
sd(NBA_test$WinShares)
sd(WinShares)

```

```{r}
#Random Forest Model

library(caret)
library(randomForest)
set.seed(1)

ctrl <- trainControl(method = "cv", number = 10)
grid_rf <- expand.grid(mtry = c(2, 4, 8, 16))

rf<- train(WinShares ~ ., data = NBA_train, importance=T, method = "rf", trControl = ctrl, tuneGrid = grid_rf)

varImp(rf)

predictionsRF <- predict(rf, NBA_test)
RMSE(predictionsRF, NBA_test$WinShares)
R2(predictionsRF, NBA_test$WinShares)
```
20 most important variables shown (out of 38)

TS.   100.00
TOV.   84.13
G      52.34
AST    47.82
VORP   43.83
USG.   43.67
PER    38.78
FT.    36.88
FG.    36.61
ORB    35.86
eFG.   34.99
TRB.   27.99
X2P.   27.75
TRB    27.65
STL.   27.00
FTr    23.84
STL    23.12
ORB.   22.95
MP     22.52
DRB    20.98


```{r}

#Lasso

library(glmnet)
set.seed(1)
WinShares = NBA$WinShares
lasso <- train(
WinShares ~. , data=NBA_train, method = "glmnet",
trControl = trainControl("cv", number = 10),
tuneGrid = expand.grid(alpha = 1, lambda = 10^seq(-3, 1, length =
100)))
lasso

coef(lasso$finalModel, lasso$bestTune$lambda)
predictionsL <- predict(lasso, NBA_test)
RMSE(predictionsL, NBA_test$WinShares)
R2(predictionsL, NBA_test$WinShares)

```


Section 2 - Predicting The NEXT Season's Winshares

Importing Data 
```{r}
setwd("D:/02Grad School/R-Dir/data")
NBA= read.csv("Seasons_stats_complete.csv",stringsAsFactors=TRUE)
str(NBA)

```
```{r}
library(dplyr)
NBA=NBA %>% 
  arrange(Year, Player) %>% 
  group_by(Player) %>% 
  arrange(Year) %>%
  #Leading WS to the following year
  mutate(WS_next_year=lead(WS),Year_Count= dplyr::row_number()) %>% 
  ungroup() %>% 
  filter(Year_Count > 1) %>% 
  filter(!is.na(WS_next_year)) %>% 
  select(Tm, Pos, Player, Year, Year_Count, Age, MP, G, PER, TS., X3PAr, FTr, ORB., TRB., DRB., STL., BLK., TOV., USG., OBPM, BPM, DBPM, VORP, FG, FGA, FG., X3P, X3PA, X3P., X2P, X2PA, X2P., eFG., FT, FTA, FT., ORB, DRB, TRB, AST, STL, BLK, WS, WS_next_year) %>% 
filter(Year >= 1999) %>% 
filter(MP >= 1003) %>% 
filter(G >= 57)

```


```{r}
#Selecting correlated numerical variables
num2<-dplyr::select_if(NBA, is.numeric)
num2

corr2<-cor(NBA$WS_next_year, num2)
corr2
correlated2= which(corr2>0)
correlated2
summary(correlated2)


```


```{r}
library(ggplot2)
#Checking Missing Cases
missing<-sum(!complete.cases(NBA))
missing

cm=colMeans(is.na(NBA))
cm
counts=which(cm>0.0)
counts
length(counts)
cm[counts]

#Assigning a variable to Win Shares
WinShares = NBA$WS

#Assigning a variable to Win Shares for the next year
WinShares_Next_Year = NBA$WS_next_year

#Scaling, Taking out Categorical variables, Year, and the Winshare_Next Year dependent variable
NBA1<- as.data.frame(scale(NBA[, -c(1, 2, 3, 4, 44)]))

#Binding the variables together
NBA = cbind(NBA1, WinShares_Next_Year)

#Histogram
histogram = ggplot(NBA, aes(x=WinShares)) + geom_histogram()
histogram

#Checking for missing cases
missing<-sum(!complete.cases(NBA))
missing


```



```{r}
#Splitting the data

set.seed(1)
library(caret)


indexTrain <- createDataPartition(NBA$WinShares_Next_Year, p = 0.9,list = FALSE)
NBA_train <- NBA[indexTrain,]
NBA_test <- NBA[-indexTrain,]

indexTrainValidation <- createDataPartition(y = NBA_train$WinShares_Next_Year, p = 0.9,list = FALSE)
NBA_val <- NBA_train[-indexTrainValidation,]
```

```{r}
#Lasso

library(glmnet)
set.seed(1)
WinShares_Next_Year = NBA$WinShares_Next_Year
lasso <- train(
WinShares_Next_Year ~. , data=NBA_train, method = "glmnet",
trControl = trainControl("cv", number = 10),
tuneGrid = expand.grid(alpha = 1, lambda = 10^seq(-3, 1, length =
100)))
lasso

coef(lasso$finalModel, lasso$bestTune$lambda)
predictionsL <- predict(lasso, NBA_test)
RMSE(predictionsL, NBA_test$WinShares_Next_Year)
R2(predictionsL, NBA_test$WinShares_Next_Year)


A = coef(lasso$finalModel, lasso$bestTune$lambda)

Var_Check = rownames(A)[A @ i]
Var_Check
```


```{r}
NBA=NBA %>%
  select("Year_Count" , "Age"  , "G" , "PER" , "TS.", "X3PAr", "FTr", "ORB.", "DRB.", "STL.", "TOV.", "USG.", "OBPM", "BPM" , "VORP", "X3P", "X3PA", "X3P." , "X2PA", "X2P.", "eFG.", "FT.", "ORB", "DRB", "BLK", "WS", "WinShares_Next_Year")

indexTrain <- createDataPartition(NBA$WinShares_Next_Year, p = 0.9,list = FALSE)
NBA_train <- NBA[indexTrain,]
NBA_test <- NBA[-indexTrain,]

indexTrainValidation <- createDataPartition(y = NBA_train$WinShares_Next_Year, p = 0.9,list = FALSE)
NBA_val <- NBA_train[-indexTrainValidation,]

```

```{r}
#Linear Regression
set.seed(1)
library(caret)
train.control = trainControl(method = "cv", number = 10)
linearRegression <- train(WinShares_Next_Year ~., data = NBA_train, method = "lm", trControl = train.control)
print(linearRegression)
summary(linearRegression)

#Checking the RMSE of the WinShares Next Year
predictionsLR= predict(linearRegression, NBA_test)
RMSE(predictionsLR, NBA_test$WinShares_Next_Year)
R2(predictionsLR, NBA_test$WinShares_Next_Year)

#Comparing to the Standard Deviation of the WinShares Next Year Column
sd(WinShares_Next_Year)

```

```{r}
#Random Forest Model

library(caret)
library(randomForest)
set.seed(1)

ctrl <- trainControl(method = "cv", number = 10)
grid_rf <- expand.grid(mtry = c(2, 4, 8, 10))

rf<- train(WinShares_Next_Year ~ ., data = NBA_train, importance=T, method = "rf", trControl = ctrl, tuneGrid = grid_rf)

varImp(rf)

predictionsRF <- predict(rf, NBA_test)
RMSE(predictionsRF, NBA_test$WinShares_Next_Year)
R2(predictionsRF, NBA_test$WinShares_Next_Year)
```
20 most important variables shown (out of 26)
        
WS          100.00
BPM          78.69
VORP         77.70
X2PA         65.15
OBPM         64.47
PER          61.40
X3PAr        53.41
Age          52.72
ORB          50.56
ORB.         49.92
DRB          49.87
DRB.         49.54
Year_Count   47.23
BLK          41.96
X3PA         39.21
X3P          37.08
TS.          35.44
eFG.         33.16
X2P.         28.97
X3P.         28.90


```{r}
#Ridge

set.seed(1)
ridge <- train(
WinShares_Next_Year ~., data = NBA_train, method = "glmnet",
trControl = trainControl("cv", number = 10),
tuneGrid = expand.grid(alpha = 0, lambda = 10^seq(-3, 1, length =
100)))

predictionsRidge <- predict(ridge,NBA_test)
RMSE(predictionsRidge, NBA_test$WinShares_Next_Year)
R2(predictionsRidge, NBA_test$WinShares_Next_Year)
```

```{r}
#Elastic Net

set.seed(1)
enet <- train(
WinShares_Next_Year ~., data = NBA_train, method = "glmnet",
trControl = trainControl("cv", number = 10),
tuneGrid = expand.grid(alpha =seq(0,1, length=10), lambda = 10^seq(-
3, 1, length = 100)))

predictionsElasticNet <- predict(enet, NBA_test)
RMSE(predictionsElasticNet, NBA_test$WinShares_Next_Year)
R2(predictionsElasticNet, NBA_test$WinShares_Next_Year)
```

```{r}
#Gradient Boost Model

set.seed(1)
gbm <- train(
WinShares_Next_Year ~., data = NBA_train, method = "gbm", preProc="nzv",
trControl = trainControl("cv", number = 10))
gbm

predictionsGBM <- predict(gbm,NBA_test)
RMSE(predictionsGBM, NBA_test$WinShares_Next_Year)
R2(predictionsGBM, NBA_test$WinShares_Next_Year)
```


```{r}
NBA_train
NBA_test
NBA_val

#Saving as a matrix
NBA_valy<-NBA_val[, c(27)]
NBA_valx<-NBA_val[, -c(27)]
NBA_valx<-as.matrix(NBA_valx)
NBA_valy<-as.matrix(NBA_valy)

NBA_testy<-NBA_test[, c(27)]
NBA_testx<-NBA_test[, -c(27)]
NBA_testx<-as.matrix(NBA_testx)
NBA_testy<-as.matrix(NBA_testy)

NBA_trainy<-NBA_train[, c(27)]
NBA_trainx<-NBA_train[, -c(27)]
NBA_trainx<-as.matrix(NBA_trainx)
NBA_trainy<-as.matrix(NBA_trainy)
```



```{r}
#ANN Model

library(keras)
set.seed(1)
model <- keras_model_sequential() %>% 
layer_dense(units = 50, activation = "relu", input_shape = dim(NBA_trainx)[2]) %>% 
layer_dropout(0.5) %>% 
layer_dense(units = 1)

model %>% 
compile(loss = "mse",
optimizer = "adam")

model %>% fit(NBA_trainx,
NBA_trainy,
batch_size=1000,
epochs = 500,
validation_data=list(NBA_valx, NBA_valy))

```

```{r}
#Using flags to find the best parameters
library(tfruns)
runs <- tuning_run("test.R",
flags = list(
nodes1 = c(50, 100, 150),
nodes2=c(50, 100, 150),
learning_rate = c(0.01, 0.05, 0.001, 0.0001),                
batch_size=c(100,200,500),
epochs=c(30,50,100),
activation1=c("relu","sigmoid","tanh"),
activation2=c("relu","sigmoid","tanh"),
dropout1=c(0.05, 0.1, 0.2, 0.5),
dropout2=c(0.05, 0.1, 0.2, 0.5)
 ),
sample = 0.002)
runs

view_run(runs$run_dir[1])
```
```{r}
#Using Best Numbers
library(keras)
set.seed(1)
model <- keras_model_sequential() %>% 
layer_dense(units =150, activation = "relu", input_shape = dim(NBA_trainx)[2]) %>% 
layer_dropout(0.5) %>% 
layer_dense(units = 1)

model %>% 
compile(loss = "mse",
optimizer = "adam")

model %>% fit(NBA_trainx,
NBA_trainy,
batch_size=1000,
epochs = 500)

#Calculating RMSE on test data
predictions=model %>% predict(NBA_testx)

rmse= function(x,y){
return((mean((x - y)^2))^0.5)
}

#RMSE for the ANN model
ann=rmse(predictions, NBA_testy)
ann

```


```{r}
#Re-creating the train data set to compare with other models
library(caret)
library(keras)
indexTrain <- createDataPartition(NBA$WinShares_Next_Year, p = 0.9,list = FALSE)
NBA_train<- NBA[indexTrain,]
NBA_test<- NBA[-indexTrain,]

NBA_train_re_y<-NBA_train[, c(27)]
NBA_train_re_x<-NBA_train[, -c(27)]
NBA_train_rex<-as.matrix(NBA_train_re_x)
NBA_train_rey<-as.matrix(NBA_train_re_y)

```
```{r}
library(keras)
set.seed(1)
model <- keras_model_sequential() %>% 
layer_dense(units = 150, activation = "relu", input_shape = dim(NBA_train_re_x)[2]) %>% 
layer_dropout(0.5) %>% 
layer_dense(units = 1)

model %>% 
compile(loss = "mse",
optimizer = "adam")

model %>% fit(NBA_train_rex,
NBA_train_rey,
batch_size=1000,
epochs = 500)

```
```{r}
#Calculating RMSE on test data
predictions=model %>% predict(NBA_testx)

rmse= function(x,y){
return((mean((x - y)^2))^0.5)
}

#RMSE for the ANN model
rmse(predictions, NBA_testy)
R2(predictions, NBA_testy)

```
Which model ( which hyper-parameter combination) resulted in the best accuracy on the
validation data?

Answer: The best numbers were at the 1st run with the following:
Metrics
loss	5.0459
val_loss	4.4683

Flags
nodes	150
batch_size	1000
activation	relu
learning_rate	0.0001
epochs	50


Does your best model still overfit?
Answer: With my best numbers, it does not overfit with a lower validation loss.

RMSE for best numbers:  [1] 2.020652

```{r}
#Comparing Resamples

compare=resamples(list(L=lasso, E=enet, G=gbm, R=ridge,  RF=rf))
summary(compare)

```



Section 3 - Future All Stars

```{r}
setwd("D:/02Grad School/R-Dir/data")
NBA= read.csv("Seasons_stats_complete.csv",stringsAsFactors=TRUE)
AllStars= read.csv("AllStarsHistory.csv",stringsAsFactors=TRUE)
```
```{r}
#Filtering for the last 20 Seasons

library(dplyr)
NBA1999=NBA %>% 
  arrange(Year, Player) %>% 
  mutate(Player =as.character(Player)) %>% 
mutate(Player = case_when(stringr:: str_detect(Player, "Yao Ming*") ~ "Yao Ming",TRUE~Player))%>%
  mutate(Player =as.character(Player)) %>% 
mutate(Player = case_when(stringr:: str_detect(Player, "Nikola V") ~ "Nikola Vucevic",TRUE~Player))%>%
   mutate(Player =as.character(Player)) %>% 
mutate(Player = case_when(stringr:: str_detect(Player, "Nikola J") ~ "Nikola Jokic",TRUE~Player))%>%
  group_by(Player) %>% 
  arrange(Year) %>%
  mutate(Year_Count= dplyr::row_number()) %>%
  filter(Year_Count > 1) %>%
  filter(Year >= 1999) %>%
  select(Player)

```

```{r}
#Filtering for all players in their first rookie season since 1999
library(dplyr)
NBA=NBA %>% 
  arrange(Year, Player) %>% 
   mutate(Player =as.character(Player)) %>% 
mutate(Player = case_when(stringr:: str_detect(Player, "Yao Ming*") ~ "Yao Ming",TRUE~Player))%>%
  mutate(Player =as.character(Player)) %>% 
mutate(Player = case_when(stringr:: str_detect(Player, "Nikola V") ~ "Nikola Vucevic",TRUE~Player))%>%
   mutate(Player =as.character(Player)) %>% 
mutate(Player = case_when(stringr:: str_detect(Player, "Nikola J") ~ "Nikola Jokic",TRUE~Player))%>%
  group_by(Player) %>% 
  arrange(Year) %>%
  mutate(Year_Count= dplyr::row_number()) %>% 
  ungroup() %>% 
   filter(Year_Count == 1) %>% 
 filter(Year >= 1999)

```
```{r}
library(tidyverse)
NBA1999R= distinct(NBA %>% inner_join(NBA1999))

```

```{r}
#Finding all of the NBA players who have been an All-Star at some point
AllStarsR= NBA1999R %>% inner_join(AllStars)

```



```{r}
#Comparing All Stars with the Top Win Shares performers
library(dplyr)
NBATopWS=NBA %>% 
  arrange(Year, Player, WS) %>% 
  group_by(Player) %>% 
  arrange(desc(WS)) %>%
  select(Player, Player, WS)

AllStars=AllStars %>% inner_join(NBA)%>% 
arrange(Year, Player, WS) %>% 
  group_by(Player) %>% 
  arrange(desc(WS))
str(AllStars$WS)

```


```{r}
library(ggplot2)
ggplot(AllStars, aes(x=Player,y=WS)) + geom_bar(stat="identity") + theme(axis.text.x = element_text(angle = 90, hjust = 1))

#Histogram
ggplot(AllStars, aes(x=WS)) + geom_histogram()

```



```{r}
library(dplyr)
AllStarPred=
  NBA %>% 
  select(Player) %>% 
  inner_join(NBA1999R, by= "Player") %>% 
  left_join(AllStarsR %>% select(Player) %>% mutate(AllStar = 1)) %>% 
  mutate(AllStar = replace_na(AllStar, 0))

  
str(AllStarPred$AllStar)


```


```{r}

#Scaling
AllStar = AllStarPred$AllStar
AllStarPred1<- as.data.frame(scale(AllStarPred[, -c(1, 2, 4, 6, 51, 52)]))


#Binding the variables together
AllStarPred = cbind(AllStarPred1, AllStar)
summary(AllStarPred$MP)

```

```{r}
#Correlation
library(dplyr)

#Selecting correlated numerical variables
num<-dplyr::select_if(AllStarPred, is.numeric)
num

corr<-cor(AllStar, num)
quantile(corr)
correlated= which(corr>0)
correlated= AllStarPred[,correlated]
str(correlated)
summary(correlated)
names(correlated)

```

```{r}
AllStarPred = AllStarPred %>%
  mutate(AllStar = factor(AllStar))
set.seed(1)
test_sample = sample(1219, 200)
AllStarPred_test = AllStarPred[test_sample,]
AllStarPred_train = AllStarPred[-test_sample,]

```

```{r}
#Lasso

library(ROCR)
library(caret)
library(glmnet)
set.seed(1)
lasso <- train(
  AllStar ~. , data=AllStarPred_train, method = "glmnet", metric="Kappa",
  trControl = trainControl("cv", number = 10),
  tuneGrid = expand.grid(alpha = 1, lambda = 10^seq(-3, 3, length =
                                                      100)))

table(AllStarPred_test$AllStar)
predict.lasso = predict(lasso, AllStarPred_test)
confusionMatrix(predict.lasso, AllStarPred_test$AllStar)

lasso_predictions_prob=predict(lasso, AllStarPred_test, type="prob")
head(lasso_predictions_prob)

pred_lasso = prediction(lasso_predictions_prob$`1`, AllStarPred_test$AllStar)
performance(pred_lasso, measure = "auc")@y.values
perf <- performance(pred_lasso, measure = "tpr", x.measure = "fpr")

#Plotting the ROC Curve
plot(perf, col = "blue")


A = coef(lasso$finalModel, lasso$bestTune$lambda)
A
rownames(A)[A @ i]


```
AUC: 0.8364055
Accuracy : 0.94 


Lasso selected these variables:
Year        -3.502742e-01
Age         -1.048057e+00
G            3.588352e-01        
X3PAr       -1.724365e-01
FTr          1.164899e-02        
TRB.         1.906031e-01
AST.         2.092446e-01      
BLK.         2.611002e-01        
USG.         1.831474e-01         
DWS          7.865672e-02
WS           3.143331e-01          
DBPM         3.421913e-06
BPM          4.996928e-01
VORP         2.027867e-01          
X3PA        -1.137637e-01
X3P.         2.279535e-01     
X2PA         7.398146e-02          
eFG.        -1.771028e-01
FT           1.972875e-01          
FT.          2.396745e-01


```{r}

AllStarPred = AllStarPred %>%
  mutate(AllStar = factor(AllStar))
set.seed(1)
test_sample = sample(1219, 200)
AllStarPred_test = AllStarPred[test_sample,]
AllStarPred_train = AllStarPred[-test_sample,]

```

```{r}
#Ridge

set.seed(1)
ridge <- train(
AllStar ~., data = AllStarPred_train, method = "glmnet", metric="Kappa",
trControl = trainControl("cv", number = 10),
tuneGrid = expand.grid(alpha = 0, lambda = 10^seq(-3, 1, length =
100)))

table(AllStarPred_test$AllStar)
predict.ridge = predict(ridge, AllStarPred_test)
confusionMatrix(predict.ridge, AllStarPred_test$AllStar)

ridge_predictions_prob=predict(ridge, AllStarPred_test, type="prob")

pred_ridge = prediction(ridge_predictions_prob$`1`, AllStarPred_test$AllStar)
performance(pred_ridge, measure = "auc")@y.values

perfR <- performance(pred_ridge, measure = "tpr", x.measure = "fpr")

#Plotting the ROC Curve
plot(perfR, col = "green")



```
AUC: 0.8379416
Accuracy : 0.935


```{r}
#Elastic Net

set.seed(1)
enet <- train(
AllStar ~., data = AllStarPred_train, method = "glmnet",
trControl = trainControl("cv", number = 10),
tuneGrid = expand.grid(alpha =seq(0,1, length=10), lambda =
10^seq(-3, 3, length = 100)))

table(AllStarPred_test$AllStar)
predict.enet= predict(enet, AllStarPred_test)
confusionMatrix(predict.enet, AllStarPred_test$AllStar)

enet_predictions_prob=predict(enet, AllStarPred_test, type="prob")

pred_enet = prediction(enet_predictions_prob$`1`, AllStarPred_test$AllStar)
performance(pred_enet, measure = "auc")@y.values


perfE <- performance(pred_enet, measure = "tpr", x.measure = "fpr")

#Plotting the ROC Curve
plot(perfE, col = "purple")


```

AUC: 0.8379416
Accuracy : 0.935



```{r}
#Using Variables from the Lasso Selection for Logistic Regression

AllStarPred=AllStarPred %>%
  select("Year",  "Age"    ,   "G", "X3PAr",   "FTr"   ,    "TRB."    ,    "AST."     ,   "BLK."    ,    "USG."  ,       "DWS"    ,     "WS"   ,     "DBPM"    ,     "BPM"     ,   
"X3PA"     ,   "X3P."      ,   "X2PA"    ,    "eFG."     ,    "FT"     ,    "FT."    )
```

```{r}
#Logistic

set.seed(1)
# str(train_sample)
model.logistic <- glm(formula=AllStar ~ ., data=AllStarPred_train, family="binomial", maxit = 100)
summary(model.logistic)
```
```{r}
table(AllStarPred_test$AllStar)
predict.logistic <- predict(model.logistic, AllStarPred_test, type="response")
predict.logistic.label = factor(ifelse(predict.logistic > .5, "Yes", "No"))
actual.label = AllStarPred_test$AllStar
table(actual.label, predict.logistic.label)


```


```{r}
library(pROC)
ROC <- roc(AllStarPred_test$AllStar, predict.logistic)

#Plotting the ROC Curve
ROCplot = plot(ROC, col = "red")

#AUC= The area under the curve
auc(ROC)

```
AUC: 0.8015
Accuracy: 0.865