SHAPforxgboost2

rm(list=ls())
setwd("F:/healthy retail")
set.seed(123)
#install.packages("SHAPforxgboost")
require(SHAPforxgboost)
#install.packages("varhandle")
library(varhandle)
data=read.csv("rdata_mRFEI.csv")
head(data)
data=data[1:3000,]

X1=as.matrix(data[which(names(data) != "mrfei")])
head(X1)
ntree=5
mod1 = xgboost::xgboost(
  data = X1, label = data$mrfei, gamma = 0, eta =1,
  lambda = 0,nrounds = ntree, verbose = FALSE)

# shap.values(model, X_dataset) returns the SHAP
# data matrix and ranked features by mean|SHAP|
shap_values<- shap.values(xgb_model = mod1, X_train = X1) #returns the SHAP
shap_score<-shap_values$shap_score #all shap scores

View(shap_score)

shap_data<-cbind.data.frame(shap_score, shap_values$BIAS0) #add intercept to the score data


pred_mod<-predict(mod1,X1,ntreelimit=ntree) #add predicted values to the score data
shap_data[, `:=`(rowSum = round(rowSums(shap_data),6), pred_mod = round(pred_mod,6))] #add row sum and predicted values


View(shap_data)
View(shap_score)
colMeans(shap_data)

shap_mean<-shap_values$mean_shap_score #mean shap scores
sm1
sm1=cbind.data.frame(lab=labels(shap_mean),sm=as.numeric(round(shap_mean,3)) ) 
ggplot(data=sm1,aes(x=reorder(lab,sm),y=sm) )+
  geom_bar(stat="identity")+
  coord_flip()

shap_long<- shap.prep(xgb_model = mod1, X_train = X1) #long format for graph
shap.plot.summary(shap_long)

sm2=rbind.data.frame(sm1,cbind.data.frame(lab="mean_pred",sm=round(mean(pred_mod),3)))
ggplot(data=sm2,aes(x=reorder(lab,sm),y=sm) )+
  geom_bar(stat="identity")+
  coord_flip()

cc=c()
for(i in 1:length(colnames(shap_score))){
  sh=as.matrix(shap_score)[,i]  
  or=c(X1[,i])
  co=cor(sh,or)
  cc=c(cc,colnames(shap_score)[[i]],co)
}
aa=data.frame(matrix(c(cc),ncol=2,byrow = TRUE))
bb=cbind.data.frame(X1=names(shap_mean),shap_mean)
sm3=merge(aa,bb,by="X1",sort=TRUE)
sm4=read.csv("varlabels.csv")
colnames(sm4)=c("X1","varlabs")
sm3=merge(sm3,sm4,by="X1",sort=TRUE,all.x=TRUE)
sm3=unfactor(sm3)

ggplot(data=sm3,aes(x=reorder(varlabs,shap_mean),y=shap_mean,fill=X2))+
  geom_bar(stat="identity")+
  coord_flip()+
  ggtitle("Direction of Association")+
  xlab("Predictor")+
  ylab("Mean |Shapley|")+
  labs(fill="Correlation")+
  scale_fill_gradient2(low="navyblue",mid="white",high="red")