Created
September 7, 2016 22:41
-
-
Save dkhurana1306/53eda3bd07e03b746adacc3937ec5f34 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
############################################################# | |
model_data = data_higgs_0_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
drop = c(ncol(model_data),ncol(model_data)-1,ncol(model_data)-2) | |
bst_0 <- xgboost(data = data.matrix(model_data[,-drop]), label = label, | |
max.depth =9, | |
eta = 0.01, | |
nround = 624, | |
nthread = 4, | |
objective = "binary:logistic") | |
model_data = data_higgs_1_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
drop = c(ncol(model_data),ncol(model_data)-1,ncol(model_data)-2) | |
bst_1 <- xgboost(data = data.matrix(model_data[,-drop]), label = label, | |
max.depth =9, | |
eta = 0.01, | |
nround = 652, | |
nthread = 4, | |
objective = "binary:logistic") | |
model_data = data_higgs_2_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
drop = c(ncol(model_data),ncol(model_data)-1,ncol(model_data)-2) | |
bst_2 <- xgboost(data = data.matrix(model_data[,-drop]), label = label, | |
max.depth =9, | |
eta = 0.01, | |
nround = 1104, | |
nthread = 4, | |
objective = "binary:logistic") | |
############################################################# | |
AMS <- function(real,pred,weight) | |
{ | |
pred_s_ind = which(pred==1) # Index of s in prediction | |
real_s_ind = which(real==1) # Index of s in actual | |
real_b_ind = which(real==0)# Index of b in actual | |
s = sum(weight[intersect(pred_s_ind,real_s_ind)]) # True positive rate | |
b = sum(weight[intersect(pred_s_ind,real_b_ind)]) # False positive rate | |
b_tau = 10 # Regulator weight | |
ans = sqrt(2*((s+b+b_tau)*log(1+s/(b+b_tau))-s)) | |
return(ans) | |
} | |
thresh = seq(0.04,0.13,0.001) | |
E= E_s = E_b = aMS = rep(0,length(thresh)) | |
for (i in (1:length(thresh))){ | |
model_data = data_higgs_0_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
pred_0 = predict(bst_0, data.matrix(model_data[,-drop])) | |
pred_0 = as.integer(pred_0 + thresh[i]) | |
model_data = data_higgs_1_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
pred_1 = predict(bst_1, data.matrix(model_data[,-drop])) | |
pred_1 = as.integer(pred_1 +thresh[i]) | |
model_data = data_higgs_2_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
pred_2 = predict(bst_2, data.matrix(model_data[,-drop])) | |
pred_2 = as.integer(pred_2 +thresh[i]) | |
pred_undef = rep(0,nrow(data_higgs_undefined)) | |
pred = rbind(c(pred_undef,pred_0,pred_1,pred_2)) | |
response = rbind(c(data_higgs_undefined$Label, | |
data_higgs_0_cleaned$Label, | |
data_higgs_1_cleaned$Label, | |
data_higgs_2_cleaned$Label)) | |
response = as.integer(response) -1 | |
weight = rbind(c(data_higgs_undefined$Weight, | |
data_higgs_0_cleaned$Weight, | |
data_higgs_1_cleaned$Weight, | |
data_higgs_2_cleaned$Weight)) | |
error = 1-sum(response == pred)/length(response) | |
ams = AMS(response,pred,weight) | |
pred_b_ind = which(pred==0) | |
pred_s_ind = which(pred==1) | |
real_s_ind = which(response==1) | |
real_b_ind = which(response==0) | |
class_error_s = 1- length(intersect(pred_s_ind,real_s_ind))/length(real_s_ind) | |
class_error_b = 1- length(intersect(pred_b_ind,real_b_ind))/length(real_b_ind) | |
E[i] =error | |
E_s[i] =class_error_s | |
E_b[i]= class_error_b | |
aMS[i] =ams | |
print(ams) | |
} | |
plot(thresh,aMS) | |
t0 = 0.15 | |
t1 = 0.15 | |
t2 = 0.15 | |
model_data = data.submission_higgs_0_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
drop = c(ncol(model_data)) | |
pred_0 = predict(bst_0, data.matrix(model_data[,-drop])) | |
pred_0 = as.integer(pred_0 + t0) | |
model_data = data.submission_higgs_1_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
drop = c(ncol(model_data)) | |
pred_1 = predict(bst_1, data.matrix(model_data[,-drop])) | |
pred_1 = as.integer(pred_1 + t1) | |
model_data = data.submission_higgs_2_cleaned_scaled | |
label = as.integer(model_data$Label) -1 | |
drop = c(ncol(model_data)) | |
pred_2 = predict(bst_2, data.matrix(model_data[,-drop])) | |
pred_2 = as.integer(pred_2 + t2) | |
pred_undef = rep(0,nrow(data.submission_higgs_undefined)) | |
pred = rbind(c(pred_undef,pred_0,pred_1,pred_2)) | |
EventId = rbind(c(data.submission_higgs_undefined$EventId, | |
data.submission_higgs_0_cleaned_scaled$EventId, | |
data.submission_higgs_1_cleaned_scaled$EventId, | |
data.submission_higgs_2_cleaned_scaled$EventId)) | |
df = as.data.frame(cbind(c(EventId),c(1:length(pred)),c(pred))) | |
names(df)[1] = "EventId" | |
names(df)[2] = "RankOrder" | |
names(df)[3] = "Class" | |
df$EventId = as.integer(df$EventId) | |
df$RankOrder= as.integer(df$RankOrder) | |
df$Class = ifelse(df$Class==0,"b","s") | |
write.csv(df, file = "submission.csv",row.names=FALSE,quote=FALSE) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment