THMMY.gr

ytest = data[,1]
predprobM1 = data[,2]
predprobM2 = data[,3]
pred_objM1 = prediction(predprobM1, ytest, label.ordering = c("0","1"))
pred_objM2 = prediction(predprobM2, ytest, label.ordering = c("0","1"))
ROCcurveM1 = performance(pred_objM1, "tpr","fpr")
ROCcurveM2 = performance(pred_objM2, "tpr","fpr")
plot(ROCcurveM1, col="blue")
par(new=TRUE)
plot(ROCcurveM2,col="red")
abline(0,1,col="green")

###--------------------------------------------------------------
#Το εμβαδόν μπορεί να βρεθεί: (πχ για Μ1)
performance(pred_objM1,"auc")
#και για M2
performance(pred_objM2,"auc") 
###-------------------------------------------------------------

>X1 = seq(min(alladata[, 1]), 4 , by = 0.1)
> X2 = seq(min(alldata[, 2]), 5, by = 0.1)
> mygrid = expand.grid(X1, X2)
> colnames(mygrid) = colnames(alladata)[1:2]

tst=data.frame(x1=4,x2=5)
predict(model,tst)

sum(data$P_M1 > 0.5 & data$Class == 1) / sum(data$Class == 1) #TPR
sum(data$P_M1 > 0.5 & data$Class == 0) / sum(data$Class == 0) #FPR
sum(data$P_M2 > 0.5 & data$Class == 1) / sum(data$Class == 1) #TPR
sum(data$P_M2 > 0.5 & data$Class == 0) / sum(data$Class == 0) #FPR

> m1 = data.frame(data[,c(1,2)])
> model_m1 <- naiveBayes(Class ~ . , data = m1)
> predprob = predict(model_m1, m1, type = "raw")
> pred_obj_m1 = prediction(predprob[,2], m1[,1], label.ordering = c("1","0"))
> roc_curve_m1 <- performance(pred_obj_m1, "tpr", "fpr")
> plot(roc_curve_m1, col = "green")
> abline(0,1, col = "blue")
> performance(pred_obj_m1, "auc")

Usage

Accuracy(y_pred, y_true)

Arguments

y_pred	Predicted labels vector, as returned by a classifier
y_true	Ground truth (correct) 0-1 labels vector

new_data <- data.frame(X1 = -2, X2 = -1.9)
predict(model, new_data)

> model <- svm(Y ~ ., data = data, type = "C-classification", kernel = "radial", gamma = 1000000, probability = TRUE)
> predict(model, new_data, probability = TRUE)
1 
1 
attr(,"probabilities")
          1         2
1 0.5471181 0.4528819
Levels: 1 2
> predict(model, new_data)
1 
2 
Levels: 1 2

pca_model <- prcomp(training, center = TRUE, scale = TRUE)
eigenvalues = pca_model$sdev^2

accuracies <- c()
pvalues = c(1:9)
 for ( p in pvalues){
	training_pc <- as.data.frame(predict(pca_model, training)[, 1:p])
	testing_pc <- as.data.frame(predict(pca_model, testing)[, 1:p]) 
	abla = knn(training_pc, testing_pc, trainingType, k = 3, prob = TRUE)
	accuracies = c(accuracies, Accuracy(abla,testingType))
}
bestp = pvalues[which.max(accuracies)] 

clusters = max.col(GMMmodel$posterior)
clusters[which(clusters==2)]=4
clusters[which(clusters==3)]=2
clusters[which(clusters==4)]=3
par(mfrow = c(2, 2)) 
plot(knndata[,1:2],col = knndata[,3]+1)
plot(GMMmodel, which = 2)
plot(knndata[,1:2],col = kmeans_model$cluster+1)
plot(knndata[,1:2],col = clusters+1)
acc_kmeans = Accuracy(y_pred = kmeans_model$cluster, y_true = knndata[,3])
acc_gmm = Accuracy(y_pred = clusters, y_true = knndata[,3])

> model <- naiveBayes(HighTraffic ~ ., data = traffic)
> predict(model, trvalue, type = "raw")
              No       Yes
[1,] 0.003984064 0.9960159

training_error = c()
training_error2 = c()
for (gamma in gammavalues) {
  svm_model = svm(y ~ ., kernel="radial", type="C-classification", data = trainingdata, gamma = gamma)
  pred = predict(svm_model, trainingdata[, c(1:2)])
  training_error = c(training_error, 1 - Accuracy(trainingdata$y, pred))
  training_error2 = c(training_error2, 1 - Accuracy(pred, trainingdata$y))
}

?axis

axis(side=1,at=c(1,20,30,50),labels=c("1975","1980","1985","1990"))