library(ggplot2) 
library(ggrepel)

dir<-"word_data" #word_data or char_data
type<-"antigen" #celltype or antigen
label<-"2" # "1" or "2"
num<-"3-5" # "1" to "10"

acc_mat<-c()
if(type=="celltype")
{
  class_num=87
}else
{
  class_num=76
}
true_class_total<-array(0, dim=class_num)
false_class_total<-array(0, dim=class_num)

for (f in 1:10) {
 result_file=paste("~/Documents/Automatic-Annotation/SVM-result/",dir,"/output_c10_test", f, "_human_",type,"_intl",label,"f",num,"gram.txt", sep="")
 label_file=paste("~/Documents/Automatic-Annotation/label_feature/c10_test",f,"_human_",type,"_intl",label,".txt",sep = "")
 test_result=read.table(file = result_file)
 test_label=read.table(file = label_file)
 test_result.df <- data.frame(test_result)
 
 #number of classes
 num_class<-dim(test_result.df)[2]-1
 true_class<-array(0, dim=c(num_class))
 false_class<-array(0, dim=c(num_class))
 false_index<-c() # record the index of false predicted samples
 true_index<-c()
 for (row in 2:dim(test_result.df)[1]) {
  row_value<-test_result.df[row,2:dim(test_result.df)[2]]
  col_ind<-which(row_value==max(row_value))
  class_ind <- test_result.df[1,col_ind+1]
  label_ind <- test_label[row-1,1]
  if(class_ind==label_ind) #label_ind:0-86, true_class[1:87]
  {
    true_class[label_ind+1]=true_class[label_ind+1]+1
    true_class_total[label_ind+1]=true_class_total[label_ind+1]+1
    true_index<-c(true_index,row-1)
  }else
  {
    false_class[label_ind+1]=false_class[label_ind+1]+1
    false_class_total[label_ind+1]=false_class_total[label_ind+1]+1
    false_index<-c(false_index,row-1) #row index of test_result.df
  }
 }
 
 #write index of false sample to file
 false_file = paste("~/Documents/Automatic-Annotation/SVM-result/",dir,"/c10_test",f,"_human_",type,"_l",label,"_",num,"gram","_falseindex",sep = "")
 write.table(false_index, file = false_file,col.names = F, row.names = F, quote = F, sep = "\n")
 true_file = paste("~/Documents/Automatic-Annotation/SVM-result/",dir,"/c10_test",f,"_human_",type,"_l",label,"_",num,"gram","_trueindex",sep = "")
 write.table(true_index, file = true_file,col.names = F, row.names = F, quote = F, sep = "\n")

  #write accuracy of each class to file
 filename = paste("~/Documents/Automatic-Annotation/SVM-result/",dir,"/c10_test",f,"_human_",type,"_l",label,"_",num,"gram","_accuracy",sep = "")
 acc<-c()
 acc<-c(acc, sum(true_class)/(sum(true_class)+sum(false_class)))
 for (i in 1:dim(true_class)) {
  if(true_class[i]==0)
    acc<-c(acc,0)
  else
    acc<-c(acc, true_class[i]/(true_class[i]+false_class[i]))
 }
 write.table(acc, file = filename,col.names = F, row.names = F, quote = F, sep = "\n")
 acc_mat<-rbind(acc_mat,acc)
}
dim(acc_mat)
#plot Mean accuracy and error bar of each class
mean<-c()
sd<-c()
for (c in 1:dim(acc_mat)[2]) {
  mean <- c(mean,mean(acc_mat[,c]))
  sd <- c(sd,sd(acc_mat[,c]))
}
mean
##name each column as class
l2_class<-c("Average")
class_file<-paste("~/Documents/Automatic-Annotation/label_feature/c10_human_",type,"_l",label,"_class.txt",sep = "")
class1<-read.table(file = class_file, sep="\t")
col.df<-data.frame(class1[,2])
colnames(col.df)<-c("V1")

l2_class<-rbind(t(l2_class), col.df)
l2_class<-cbind(l2_class,mean)
l2_class<-cbind(l2_class,sd)

colnames(l2_class) <- c("Class", "MeanAccuracy", "SD")
dir_out = paste("~/Documents/Automatic-Annotation/SVM-result/",dir,"/",sep = "")
fo_img = sprintf("%sc10_%s_l%s_class_%sgram_acc.pdf", dir_out, type, label,num)
t=paste("Average accuracy of each class (",dim(class1)[1]," classes in Label ",label,") for human ",type," data
        Total average is ", l2_class[1,2],"(word, ",num,"gram)",sep = "")

pdf(fo_img, width=20, height=9, pointsize=9)
g <- ggplot(l2_class, aes(x=Class, y=MeanAccuracy, colour=Class)) + 
  geom_errorbar(aes(ymin=MeanAccuracy-SD, ymax=MeanAccuracy+SD), width=.5) +
  geom_line() + geom_point() + 
  theme(axis.text.x = element_text(angle = 90, hjust = 1), 
        plot.title = element_text(family = 'Helvetica', 
        color = '#666666', face = 'bold', size = 16, hjust = 0.5))+labs(title = t)
plot(g)
dev.off()




#extract bad cases
filename=paste("~/Documents/Automatic-Annotation/Analyze/human_",type,"_l",label,"_",num,"gram","_badcase08_name.txt",sep = "")
l2_class[,2]
ind<-which(l2_class[,2]<0.8)
ind
badcase<-l2_class[ind,1]
badcase
vec_badcase<-as.vector(badcase)
write(vec_badcase,file=filename,sep = "\n")
badcase
class1[,2]
bad_class_id<-c()
for (i in 1:length(vec_badcase)) {
  id<-which(class1[,2]==vec_badcase[i])
  bad_class_id<-c(bad_class_id,id-1)
}
bad_class_id #exactly class ID (0-86)

false_class_total[id+1]
true_class_total
acc<-c()
for (i in 1:length(true_class_total)) {
  class<-true_class_total[i]/(false_class_total[i]+true_class_total[i])
  acc<-c(acc,class)
}
id<-which(acc<0.8)
 
sum(false_class_total)
sum(true_class_total)