gistfile1.txt

install.packages("tree")
library(MASS)
library(tree)
head(Pima.tr)

#############

set.seed(1234)
# enter your code here
# fit a tree in the training set
basic_tree <- tree(type~., data=Pima.tr) # fill in the ***

# use "basic tree" model above to predict in the test set
predict_basic_tree <- predict(
    basic_tree, newdata = Pima.te, type = "class")

# produce a confusion matrix showing accuracy and error
cat("\n By the way, let's obtain the test set error\n\n")
table(Pima.te$type, predict_basic_tree)
# 170 true Neg are actually Neg
# 65 true Pos are actually Pos
# the rest are errors, mistakes (44 + 53)
## this is accuracy, or error for the test set error
## with a simple tree (not cross-validated)


accuracy <- (170+65) / (170 + 53 + 44 + 65)
error_rate <- 1 - accuracy

accuracy
error_rate

######################################################
cat("\n By the way, let's obtain the training set error\n\n")
# again, this is for the simple tree (not cross-validated)
predict_basic_tree.tr <- predict(
    basic_tree, newdata = Pima.tr, type = "class")

summary(predict_basic_tree.tr)
table(Pima.tr$type, predict_basic_tree.tr)
# 122 + 55 are correctly classified
# 13 + 10 are mistakenly classified
accuracy.tr <- (122 + 55) / (122 + 55 + 23)
plot(basic_tree)
text(basic_tree)
summary(basic_tree)

########################

cat("\nThe estimated test set error rate for the basic tree (before cross-validating) is:", round(error_rate*100), "%\n\n")

cat("\nThe estimated training set error rate for the basic tree (before cross-validating) is:", round((1 - accuracy.tr)*100), "%\n\n")


###############

## Now we're going to figure out how much we should prune the tree...

## Pruned tree ##
set.seed(234235)
pruned_tree <- cv.tree(
    basic_tree, FUN=prune.misclass) 

print(pruned_tree)
# enter your code here

# let's plot the pruning result
plot(x = pruned_tree$size, y = pruned_tree$dev,
     main = "Relationship between Tree Size and Error",
    type = "b")

# label the points with value of alpha (also known as k)
text(x = pruned_tree$size, y = pruned_tree$dev,
     labels = round(pruned_tree$k, 2), 
     pos = 4, col = "darkred", cex = 1.2)

#########################################

# you now have information about how much to grow the tree
# so refit your training set: fill in the "***"
# where *** = the optimal number of nodes
best_tree <- prune.misclass(basic_tree, best = 11)

summary(best_tree)
plot(best_tree)
text(best_tree)

#######################

# and now, bring that model to your test set.
# produce another confusion matrix
predict_best_tree <- predict(best_tree, newdata = Pima.te, type = "class")
plot(best_tree)
text(best_tree)
# produce a confusion matrix showing accuracy and error
cat("\nHere is a test set accuracy rate for the cross-validated tree with 11 terminal nodes\n")
table(Pima.te$type, predict_best_tree)

accuracy_rate_1 = (173 + 72) / (173 + 50 + 37 + 72)
error_rate_1 = 1 - accuracy_rate_1

cat("\nThe test error rate for our cross-validated tree =", round(100*(1-accuracy_rate_1)), "%\n")

########################