n <- nrow(covid)
c3 <- sample(1:n, 0.8*n)
length(c3)
c4 <- (1:n)[-c3]
#c3=1:52;c4=53:66; # 80,20 training testing
d2 <- covid[c3,];d3 <- covid[c4,]

c3=1:288;c4=289:360; # 80,20 training testing

C1 <- as.matrix(covid[,1]);C3 <- as.matrix(covid[,2])
C4 <- as.matrix(covid[,3])#;C3 <- as.matrix(covid[,7])
Rt <- as.matrix(covid[,4]);R1 <- as.matrix(covid[,5])
R2 <- as.matrix(covid[,6]);R3 <- as.matrix(covid[,7])

all_data <- data.frame(C1,C3,C4,Rt,R1,R2,R3)#,R1,R2,R3)
dim(all_data)
all_data <- zoo(all_data, seq(from = as.Date("2020-01-22"), 
                              to = as.Date("2021-01-15"), by =1))
#covid <- zoo(covid, seq(from = as.Date("2020-01-22"), 
#                       to = as.Date("2020-06-09"), by =1))
tail(all_data)
dim(covid)
ytrain <- covid[c3,1]
ytest <- covid[c4,1]
head(ytest)
traindata <- all_data[c3,] # take training part
testdata <- all_data[c4,] # take test data

# ANN for training

ann_trn_fit <- monmlp.fit(x = as.matrix(traindata),
                          y = as.matrix(ytrain),
                          hidden1 = 4,
                          hidden2 = 8,
                          #monotone = 2,
                          #n.ensemble = 10,
                          #bag=TRUE,
                          iter.max  = 10000,
                          To = linear)

ann_trn_pred <- monmlp.predict(x = as.matrix(traindata),weights = ann_trn_fit)
ann_trn_pred <- ts(ann_trn_pred);accuracy(ann_trn_pred, ytrain)
ann_trn_pred <- data.frame(ann_trn_pred)

ann_test_fit <- monmlp.fit(x = as.matrix(traindata),
                           y = as.matrix(ytrain),
                           hidden1 = 4,
                           hidden2 = 8,
                           #monotone = 2,
                           #n.ensemble = 10,
                           #bag=TRUE,
                           iter.max  = 10000,
                           To = linear)

ann_test_pred <- monmlp.predict(x = as.matrix(testdata),weights = ann_test_fit)
ann_test_pred <- ts(ann_test_pred);accuracy(ann_test_pred, ytest)
##############

svm_trn <- svm(ytrain ~. , data = traindata)
#tune_par <- tune.svm(ytrain ~. , data = traindata, gamma = 10^(-#5:-1), cost = 10^(-3:1)
svm_trn_pred <- ts(svm_trn_pred, start=(2020-01-22))
svm_trn_pred <- predict(svm_trn,newdata= traindata);accuracy(svm_trn_pred,ytrain)

## SVM for testing
svm_test <- svm(ytrain ~. , data = traindata);svm_test_pred <- ts(svm_test_pred, start=(2020-11-05))
##tune_par <- tune.rf(ytrain ~. , data = traindata, gamma = 10^(-#5:-1), cost = 10^(-3:1)
svm_test_pred <- predict(svm_test,newdata= testdata);accuracy(svm_test_pred,ytest)

############################################
# RF for training
rf_trn <- randomForest(ytrain ~., data = traindata, keep.forest = TRUE) 
rf_trn_pred <- predict(rf_trn, traindata);accuracy(rf_trn_pred,ytrain)

rf_trn_pred <- ts(rf_trn_pred,start(2020-01-22), end(2020-11-04),frequency = 1/7)

# RF for testing
rf_test <- randomForest(ytrain ~., data = traindata, keep.forest = TRUE) 

rf_test_pred <- predict(rf_test, testdata);accuracy(rf_test_pred,ytest)

rf_test_pred <- ts(rf_test_pred,start(2020-11-05), end(2021-01-16),frequency = 1/7)
## tuing rf
# mtry <- sqrt(ncol(data))
#control <- trainControl(method='repeatedcv', 
#                        number=5, 
#                       repeats=3,
#                        search = 'random')
#
#Random generate  mtry values with tuneLength = 15
#set.seed(1)
#rf_random <- train(Class ~ .,
#                   data = dataset,
#                   method = 'rf',
#                   metric = 'Accuracy',
#                   tuneLength  = 15, 
#                   trControl = control)
#print(rf_random)
##########################################
# KNN for training
knn_trn <- knnreg(ytrain~.,traindata,k=7);knn_trn_pred <- predict(knn_trn, traindata)

knn_trn_pred <- ts(knn_trn_pred, start(2020-01-22));accuracy(knn_trn_pred, ytrain)

# KNN for testing
knn_test <- knnreg(ytrain~.,traindata,k=7);knn_test_pred <- predict(knn_test, testdata)
#knn.cross <- tune.knn(x = full.data, y = full.dir, k = #1:20,tunecontrol=tune.control(sampling = "cross"), cross=5)

knn_test_pred <- ts(knn_test_pred, start(2020-11-05));accuracy(knn_test_pred, ytest)