```{r}
library(bnlearn)
library(gRain)


prediction <- function(junctree, data, observed_var, predicted_var) {
predictions <- NULL
X <- NULL
  #loop through rows in data
for (i in 1:nrow(data)) {
#loop through observed variables and set vector X to corresponding value
  for (j in observed_var) {
X[j] <- if(data[i, j] == "yes") "yes" else "no" }
#Get the list of potentials
evidence <- setFinding(object = junctree, nodes = observed_var,states = X)
#Get the probabilities for the values of the predicted value
prob <- querygrain(object = evidence,nodes = predicted_var)$S
predictions[i] <- if (prob["yes"] > 0.5) "yes" else "no"

}
return (predictions)
}
```
```{r}
set.seed(567)
data("asia")
ind <- sample(1:5000, 4000)
tr <- asia[ind,]
te <- asia[-ind,]

#Create the native bayesian network
BN <-model2network("[S][A|S][T|S][L|S][B|S][E|S][X|S][D|S]")

trainingPoints <- c(10,20,50,100,1000,2000)

accuracys <- matrix(NA, length(trainingPoints))

for (i in 1:length(trainingPoints)) {

trainData <- tr[1:trainingPoints[i],]

#Fit parameters of network to train data
BN_fit <- bn.fit(x = BN, data = trainData, method="bayes")

#Convert to gRain-object
BN_grain <- as.grain(BN_fit)

#Get junctiontree
junctree <- compile(BN_grain)

# #Make predictions
pred <- prediction(junctree = junctree, data = te, observed_var = c("A", "T", "L", "B", "E", "X", "D"),predicted_var = c("S"))

# Calculate confusion matricies
confusion_matrix<- table(pred, te$S)
accuracys[i] <- sum(diag(confusion_matrix))/ sum(confusion_matrix)
}
print(accuracys)

```
```{r}
#Create the native bayesian network
BN <-model2network("[A][T][L][B][E][X][D][S|A:T:L:B:E:X:D]")

trainingPoints <- c(10,20,50,100,1000,2000)

accuracys <- matrix(NA, length(trainingPoints))

for (i in 1:length(trainingPoints)) {

trainData <- tr[1:trainingPoints[i],]

#Fit parameters of network to train data
BN_fit <- bn.fit(x = BN, data = trainData, method="bayes")

#Convert to gRain-object
BN_grain <- as.grain(BN_fit)

#Get junctiontree
junctree_naive <- compile(BN_grain)

# #Make predictions
pred <- prediction(junctree = junctree_naive, data = te, observed_var = c("A", "T", "L", "B", "E", "X", "D"),predicted_var = c("S"))

# Calculate confusion matricies
confusion_matrix<- table(pred, te$S)

accuracys[i] <- sum(diag(confusion_matrix))/ sum(confusion_matrix)
}
print(accuracys)
```