-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathMasterFile.R
310 lines (208 loc) · 11 KB
/
MasterFile.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
setwd("~/Group05")
set.seed(1)
# Load in packages
library(tm) #general text mining package
library(SnowballC) #package for stemming
library(dplyr) #data analysis package, also loads in magrittr for %>% notation
# Load in user functions for analysis
source('FunctionsFile.R')
# Store email length
hil <- read.delim("HRC_train.tsv", sep = "", header = F)
colnames(hil) <- c("sender", "email")
email_length <- nchar(as.character(hil$email))
# We separate the vectors in the emails data frame
# for use in the model estimations
senders <- hil$sender
emails <- hil$email
# Run Preprocessing function
test.filepath <- "HRC_test.tsv"
train.filepath <- "HRC_train.tsv"
hil.test <- preprocessing(test.filepath,
ShowTerms = F,
RemoveCommonWords = F) %>%
as.data.frame()
hil.train <- preprocessing(train.filepath,
ShowTerms = T,
RemoveCommonWords = F) %>%
as.data.frame()
# Only include columns in test and training set
# for terms that appear in both document term matrices
IntersectFiles <- FindIntersect(test_emails = hil.test,
train_emails = hil.train,
return_which = "both")
hil.train <- IntersectFiles$train
hil.test <- IntersectFiles$test
rm(IntersectFiles)
# Append power features
hil.train <- AddPowerFeatures(hil.train)
hil.test <- AddPowerFeatures(hil.test)
##############################################################
###### Building A Supervised Classifier: Random Forests ######
##############################################################
library(randomForest)
# Since the ideal subset size for random forests for a classification
# is the square root of the total number of predictors,
# we will do so for this problem
mpred <- (floor(sqrt(ncol(dtm.data))))
##### INSERT FUNCTION THAT APPLIES FitRFClean FUNCTION TO EACH WORD VECTOR #####
rf.full.model <- FitRFClean(words_of_interest = names(hil.train),
train_emails = hil.train,
senders = senders)
rf.full.model.200 <- FitRFClean(words_of_interest = names(hil.train),
train_emails = hil.train,
senders = senders,
ntree = 200)
rf.full.model.300 <- FitRFClean(words_of_interest = names(hil.train),
train_emails = hil.train,
senders = senders,
ntree = 300)
importance.200 <- data.frame(importance(rf.full.model.200[[1]]),
word = rownames(importance(rf.full.model.200[[1]]))) %>%
arrange(desc(MeanDecreaseGini))
important.words.200 <- as.character(importance.200$word)
importance.100 <- data.frame(importance(rf.full.model[[1]]),
word = rownames(importance(rf.full.model[[1]]))) %>%
arrange(desc(MeanDecreaseGini))
important.words.100 <- as.character(importance.100$word)
importance.300 <- data.frame(importance(rf.full.model.300[[1]]),
word = rownames(importance(rf.full.model.300[[1]]))) %>%
arrange(desc(MeanDecreaseGini))
important.words.300 <- as.character(importance.300$word)
rf.200.1000 <- FitRFClean(words_of_interest = important.words.200[1:1000],
train_emails = hil.train,
senders = senders,
ntree = 200)
rf.200.2000 <- FitRFClean(words_of_interest = important.words.200[1:2000],
train_emails = hil.train,
senders = senders,
ntree = 200)
rf.200.3000 <- FitRFClean(words_of_interest = important.words.200[1:3000],
train_emails = hil.train,
senders = senders,
ntree = 200)
rf.200.4000 <- FitRFClean(words_of_interest = important.words.200[1:3000],
train_emails = hil.train,
senders = senders,
ntree = 200)
rf.200.5000 <- FitRFClean(words_of_interest = important.words.200[1:3000],
train_emails = hil.train,
senders = senders,
ntree = 200)
rf.100.1000 <- FitRFClean(words_of_interest = important.words.100[1:1000],
train_emails = hil.train,
senders = senders,
ntree = 100)
rf.100.2000 <- FitRFClean(words_of_interest = important.words.100[1:2000],
train_emails = hil.train,
senders = senders,
ntree = 100)
rf.100.3000 <- FitRFClean(words_of_interest = important.words.100[1:3000],
train_emails = hil.train,
senders = senders,
ntree = 100)
rf.100.4000 <- FitRFClean(words_of_interest = important.words.100[1:4000],
train_emails = hil.train,
senders = senders,
ntree = 100)
rf.100.5000 <- FitRFClean(words_of_interest = important.words.100[1:5000],
train_emails = hil.train,
senders = senders,
ntree = 100)
rf.300.1000 <- FitRFClean(words_of_interest = important.words.300[1:1000],
train_emails = hil.train,
senders = senders,
ntree = 300)
rf.300.2000 <- FitRFClean(words_of_interest = important.words.300[1:2000],
train_emails = hil.train,
senders = senders,
ntree = 300)
rf.300.3000 <- FitRFClean(words_of_interest = important.words.300[1:3000],
train_emails = hil.train,
senders = senders,
ntree = 300)
rf.300.4000 <- FitRFClean(words_of_interest = important.words.300[1:4000],
train_emails = hil.train,
senders = senders,
ntree = 300)
rf.300.5000 <- FitRFClean(words_of_interest = important.words.300[1:5000],
train_emails = hil.train,
senders = senders,
ntree = 300)
rf_tree_errors <- list(rf.100.1000, rf.100.2000, rf.100.3000, rf.100.4000, rf.100.5000, rf.200.1000, rf.200.2000, rf.200.3000, rf.200.4000, rf.200.5000, rf.300.1000, rf.300.2000, rf.300.3000, rf.300.4000, rf.300.5000, rf.full.model, rf.full.model.200, rf.full.model.300)
rf_oobs <- sapply(rf_tree_errors, function(x) x[[2]])
rf_ntree <- as.factor(sapply(rf_tree_errors, function(x) x[[1]]$ntree))
rf_words <- c(1000, 2000, 3000, 4000, 5000, 1000, 2000, 3000, 4000, 5000,1000, 2000, 3000, 4000, 5000, 8878, 8878, 8878)
rf_plotdata <- data.frame(rf_words, rf_oobs, rf_ntree) %>%
group_by(rf_words) %>%
arrange(rf_words)
ggplot(data = rf_plotdata, aes(x = rf_words, y = rf_oobs, col = rf_ntree)) +
geom_line() + geom_point() + coord_cartesian(xlim = c(0, 10000), ylim = c(0.24, 0.28))+
scale_x_discrete(limits = seq(from = 0, to = 9000, by = 1000)) +
scale_color_manual(values = c("#3342FF", "#BB0600", "#5BBB00")) +
labs(title = "Random Forest Out of Bound Error Plot",
x = "Power Words", y = "OOB Error", color = "Tree Size")
## put in the other models here
RF.test.predictions <- predict(rf.200.1000[[1]], hil.test)
RF.test.predictions <- TestPrediction(rf.200.1000, hil.test)
export_txt(RF.test.predictions, "predict")
############ SVM ##################
error_vector<-rep(0,0)
words1800<-importance.200 %>%
group_by(word) %>%
arrange(desc(MeanDecreaseGini))%>%
select(word)%>%
head(1700)
intersect(names(hil.train), words1800$word)
subset.hil.train<-hil.train[, intersect(names(hil.train), words1800$word)]
subset.hil.train<-cbind(senders,subset.hil.train)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "linear",
tunecontrol = tune.control(cross=5),cost = 0.00001, scale = FALSE)
error_vector<-c(error_vector,tune.fit$performances$error)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "linear",
tunecontrol = tune.control(cross=5),cost = 0.0001, scale = FALSE)
error_vector<-c(error_vector,tune.fit$performances$error)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "linear",
tunecontrol = tune.control(cross=5),cost = 0.00012, scale = FALSE)
error_vector<-c(error_vector,tune.fit$performances$error)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "linear",
tunecontrol = tune.control(cross=5),cost = 0.00011, scale = FALSE)
error_vector<-c(error_vector,tune.fit$performances$error)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "linear",
tunecontrol = tune.control(cross=5),cost = 0.0000999, scale = FALSE)
error_vector<-c(error_vector,tune.fit$performances$error)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "polynomial",
tunecontrol = tune.control(cross=5),cost = 0.1, scale = FALSE,degree =2)
error_vector<-c(error_vector,tune.fit$performances$error)
tune.fit<-tune(svm, subset.hil.train,subset.hil.train[,1], kernel = "radial",
tunecontrol = tune.control(cross=5),cost = 0.1, scale = FALSE,gamma = 1)
error_vector<-c(error_vector,tune.fit$performances$error)
svm.fit<-svm(hil.train,hil.train[,1], kernel = "linear",
tunecontrol = tune.control(cross=5),cost = 0.0000999, scale = FALSE)
q_pred<-predict(svm.fit, hil.train)
##############################################################
##### K-Means Cluster Analysis Using 100 important terms #####
# Load dependencies
library(gtools)
# Conduct kmeans clustering, K = 5 clearly
# First, extract top 100 features from random forests
top.100 <- data.frame(importance(rf.200.1000[[1]]),
word = rownames(importance(rf.200.1000[[1]]))) %>%
arrange(desc(MeanDecreaseGini))
top.100vec <- head(as.character(top.100$word), 100)
top.100.no.char <- head(as.character(top.100$word), 101)[-which(top.100$word == "email_chars")]
WFM <- hil.train[, colnames(hil.train) %in% top.100vec]
word.kmean <- kmeans(WFM , centers = 5, iter.max = 50, nstart = 50)
first_guess <- LowErrorRate(word.kmean, y = senders)
WFM_nochar <- hil.train[, colnames(hil.train) %in% top.100.no.char]
for(i in 1:nrow(WFM_nochar)){
WFM_nochar[i,] = WFM_nochar[i,] / hil.train$email_chars[i]
}
nochar.kmean <- kmeans(WFM_nochar, centers = 5, iter.max = 50, nstart = 50)
normalized_guess <- LowErrorRate(nochar.kmean, y = senders)
WFM_standard <- hil.train[, colnames(hil.train) %in% top.100vec] %>%
apply(2, function(x){
x = (x - mean(x)) / sd(x)
})
standard.kmean <- kmeans(WFM_standard, centers = 5, iter.max = 50, nstart = 50)
standard.guess <- LowErrorRate(standard.kmean, y = senders)
word.kmean <- kmeans(WFM , centers = 5, iter.max = 50, nstart = 50)