spam.data = read.table("spam.data")
dim(spam.data)
#[1] 4601   58
test = scan("spam.traintest")
sum(test)
#1536
> names(spam.data)
#[1] "V1"  "V2"  "V3"  "V4"  "V5"  "V6"  "V7"  "V8"  "V9"  "V10" "V11" "V12"
#[13] "V13" "V14" "V15" "V16" "V17" "V18" "V19" "V20" "V21" "V22" "V23" "V24"
#[25] "V25" "V26" "V27" "V28" "V29" "V30" "V31" "V32" "V33" "V34" "V35" "V36"
#[37] "V37" "V38" "V39" "V40" "V41" "V42" "V43" "V44" "V45" "V46" "V47" "V48"
#[49] "V49" "V50" "V51" "V52" "V53" "V54" "V55" "V56" "V57" "V58"

# For reasons I really don't understand, if I assign names to the database, then I can't fit the tree model (see below).  Suggestions welcome.

spam.data$V58 = factor(spam.data$V58) # this will give a classification tree
library(MASS)
library(rpart)

spam.tree = rpart(V58 ~ . data = spam.data[test==0,], cp = 0.001)
printcp(spam.tree)
#output omitted, it's long (59 splits)

attributes(spam.tree)  # this is helpful to avoid printing the whole tree
#$names
# [1] "frame"     "where"     "call"      "terms"     "cptable"   "splits"   
# [7] "method"    "parms"     "control"   "functions" "y"         "ordered"  

#$class
#[1] "rpart"


attributes(spam.tree$cptable)
#$dim
#[1] 43  5

#$dimnames
#$dimnames[[1]]
# [1] "1"  "2"  "3"  "4"  "5"  "6"  "7"  "8"  "9"  "10" "11" "12" "13" "14"
#[15] "15" "16" "17" "18" "19" "20" "21" "22" "23" "24" "25" "26" "27" "28"
#[29] "29" "30" "31" "32" "33" "34" "35" "36" "37" "38" "39" "40" "41" "42"
#[43] "43"

#$dimnames[[2]]
#[1] "CP"        "nsplit"    "rel error" "xerror"    "xstd"     

min(spam.tree$cptable[,4])   # this is xerror
# [1] 0.2998988

> .299+.016  # adding on 1 standard error
# [1] 0.315
# this would lead to a tree with 28 splits and cp = . 0.0027259

# the book claimed 17 terminal nodes was best, and printcp(spam.tree)
# seems to indicate that cp = 0.0043 will do that

spam.tree2 = rpart(V58 ~ ., data=spam.data[test==0,],cp=.0043)

# but this has 10 splits and 11 terminal nodes
spam.tree2$cptable[,2]
# 1  2  3  4  5  6  7  8  9 
# 0  1  2  4  5  6  7  8 10 
 
# and slightly worse predictions that in Table 9.3
table(predict(spam.tree2,spam.data[test==1,],type="class"),
      spam.data[test==1,58])/sum(test)   

   
#             0          1
#  0 0.57942708 0.06184896
#  1 0.03320312 0.32552083


# but here's one that's a little better (yay!)

spam.tree3 = rpart(V58 ~ ., data=spam.data[test==0,],cp=.0025)
table(predict(spam.tree3,spam.data[test==1,],type="class"),
      spam.data[test==1,58])/sum(test)
      
#             0          1
#  0 0.58072917 0.05468750
#  1 0.03190104 0.33268229

# and finally this gives a pretty nice looking tree

plot(spam.tree3, uniform = T); text(spam.tree3, use.n=T, cex=0.8)

## sad that I can't get the names on: here is what I tried

names(spam.data) = spam.names  # spam.names read in from the repository

names(spam.data)
# [1] "make"       "address"    "all"        "3d"         "our"       
# [6] "over"       "remove"     "internet"   "order"      "mail"      
#[11] "receive"    "will"       "people"     "report"     "addresses" 
#[16] "free"       "business"   "email"      "you"        "credit"    
#[21] "your"       "font"       "000"        "money"      "hp"        
#[26] "hpl"        "george"     "650"        "lab"        "labs"      
#[31] "telnet"     "857"        "data"       "415"        "85"        
#[36] "technology" "1999"       "parts"      "pm"         "direct"    
#[41] "cs"         "meeting"    "original"   "project"    "re"        
#[46] "edu"        "table"      "conference" ";"          "("         
#[51] "["          "!"          "$"          "#"          "average"   
#[56] "longest"    "total"      NA          

names(spam.data)[58]= "email"

rpart(email ~ ., data = spam.data, method="class", cp = 0.0025)
# Error in `[.data.frame`(frame, predictors) : undefined columns selected