-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathExperiment1.R
198 lines (163 loc) · 8.22 KB
/
Experiment1.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
setwd("~/Documents/013-14/Summer 2014/MedIX/ResourceAllocation")
## Loading add-on packages
pack.names <- c("rpart","rpart.plot", "pROC")
sapply(pack.names,library,character.only=TRUE)
##Formula for decision tree with all image features
#Code for img_fs with combined gabor features
formula = as.formula("label ~ markov1 + markov2 + markov3 + markov4 + markov5 +
SDIntensityBG + IntensityDifference + avg.gabor.mean + avg.gabor.SD + Energy + Homogeneity + Entropy +
thirdordermoment + Inversevariance + Sumaverage + Variance + Clustertendency + MaxProbability +
Circularity + Compactness + Eccentricity + Solidity + Extent + RadialDistanceSD + SecondMoment +
Area + ConvexArea + Perimeter + ConvexPerimeter + EquivDiameter + MajorAxisLength +
MinorAxisLength")
label.selector <- function(x,index)
{
x <- as.vector(t(x))
value <- x[seq(0,4*(length(index)-1),4)+index]
return(value)
}
mode <- function(x)
{
tabSmpl<-tabulate(x)
ifelse((sum(tabSmpl == max(tabSmpl))>1),ceiling(mean(x)),which(tabSmpl== max(tabSmpl)))
}
rescale <- function(x)
{
value <- ifelse(x==1|x==2,1,ifelse(x==3,2,3))
return(value)
}
## Loading the data
data <- read.csv("LIDC dataset with full annotations.csv",header=TRUE)
gabor.features <- data[, 19:42]
avg.gabor.mean <- (gabor.features$gabormean_0_0 + gabor.features$gabormean_0_1 + gabor.features$gabormean_0_2 +
gabor.features$gabormean_1_0 + gabor.features$gabormean_1_1 + gabor.features$gabormean_1_2 +
gabor.features$gabormean_2_0 + gabor.features$gabormean_2_1 + gabor.features$gabormean_2_2 +
gabor.features$gabormean_3_0 + gabor.features$gabormean_3_1 + gabor.features$gabormean_3_2)/12
avg.gabor.SD <- (gabor.features$gaborSD_0_0 + gabor.features$gaborSD_0_1 + gabor.features$gaborSD_0_2
+ gabor.features$gaborSD_1_0 + gabor.features$gaborSD_1_1 + gabor.features$gaborSD_1_2
+ gabor.features$gaborSD_2_0 + gabor.features$gaborSD_2_1 + gabor.features$gaborSD_2_2
+ gabor.features$gaborSD_3_0 + gabor.features$gaborSD_3_1 + gabor.features$gaborSD_3_2)/12
avg.gabor.features <- data.frame(avg.gabor.mean, avg.gabor.SD)
img_fs <- data[,c(5:18, 43:69)]
img_fs <- data.frame(img_fs, avg.gabor.features)
##Process labels
#currently iterative labeling for both trail and test
labels <- data[,70:73]
#shuffles labels
labels <- t(apply(labels,1,sample))
#takes the mode for each iteration
labels <- cbind(labels[,1],apply(labels[,1:2],1,mode),
apply(labels[,1:3],1,mode),apply(labels,1,mode))
labels <- apply(labels,c(1,2),rescale)
## Label tracker
label.tracker <- rep(1,nrow(labels))
##Balance
ones <- which(labels[,4]==1) #201 24.8%
twos <- which(labels[,4]==2) #341 42.1%
threes <- which(labels[,4]==3) #268 33.1%
#twos will be slightly undersampled
#so that they don't represent more than
#40% of the cases or 324 total
#793 is new total case number
##Stratify 60% training, 30% testing, 10% validation
train.ones <- sample(201, 121, replace=FALSE)
train.twos <- sample(341, 195, replace=FALSE)
train.threes <- sample(268, 161, replace=FALSE)
train.index <- c(ones[train.ones], twos[train.twos], threes[train.threes])
test.ones <- sample(seq(1:201)[-train.ones], 60, replace=FALSE)
test.twos <- sample(seq(1:341)[-train.twos], 97, replace=FALSE)
test.threes <- sample(seq(1:268)[-train.threes], 80, replace=FALSE)
test.index <- c(ones[test.ones], twos[test.twos], threes[test.threes])
valid.ones <- sample(seq(1:201)[-c(train.ones, test.ones)], 20,replace=FALSE)
valid.twos <- sample(seq(1:341)[-c(train.twos, test.twos)], 32, replace=FALSE)
valid.threes <- sample(seq(1:268)[-c(train.threes, test.threes)], 27, replace=FALSE)
valid.index <- c(ones[valid.ones], twos[valid.twos], threes[valid.threes])
#Finally create image feature sets
train.img <- as.matrix(img_fs[train.index,])
test.img <- as.matrix(img_fs[test.index,])
valid.img <- as.matrix(img_fs[valid.index,])
#Create consensus label vectors
#Probably redundant code with the balanced/stratified sampling code
#Train, test and valid appear in 1, 2, 3 order
cons.label <- label.selector(labels, rep(4,nrow(labels)))
train.cons.label <- cons.label[train.index]
test.cons.label <- cons.label[test.index]
valid.cons.label <- cons.label[valid.index]
cons.used.label <- c(train.cons.label, test.cons.label, valid.cons.label)
total.iter.miss <- vector(mode="list",length=4)
total.cons.miss <- vector(mode="list",length=4)
train.cons.miss <- vector(mode="list",length=4)
train.iter.miss <- vector(mode="list",length=4)
test.cons.miss <- vector(mode="list",length=4)
test.iter.miss <- vector(mode="list",length=4)
valid.cons.miss <- vector(mode="list",length=4)
valid.iter.miss <- vector(mode="list",length=4)
##Iterations
for(r in 1:4)
{
#Different iterative label vector for each iteration
iter.label <- label.selector(labels,label.tracker)
train.iter.label <- iter.label[train.index]
test.iter.label <- iter.label[test.index]
valid.iter.label <- iter.label[valid.index]
iter.used.label <- c(train.iter.label, test.iter.label, valid.iter.label)
#Make dataframes work for decision trees
train.data <- data.frame(cbind(train.iter.label, train.img))
colnames(train.data)[1] <- "label"
test.data <- data.frame(cbind(test.iter.label, test.img))
colnames(test.data)[1] <- "label"
valid.data <- data.frame(cbind(valid.iter.label, valid.img))
colnames(valid.data)[1] <- "label"
#THIS IS WHERE CLASSIFICATION ACTUALLY HAPPENS
train.cl.model <- rpart(formula, method = "class", data = train.data)
train.pred.label <- unlist(predict(train.cl.model, train.data, type="class"))
test.pred.label <- unlist(predict(train.cl.model, test.data, type="class"))
valid.pred.label <- unlist(predict(train.cl.model, valid.data, type="class"))
#Store predicted labels
pred.label <- c(train.pred.label, test.pred.label, valid.pred.label)
pred.label <- unlist(pred.label)
##Calculate way too many miss indices
miss.iter <- which(pred.label!= iter.used.label)
miss.cons <- which(pred.label!= cons.used.label)
total.iter.miss[r] <-length(miss.iter)
total.cons.miss[r] <-length(miss.cons)
train.cons.miss[r] <- length(which(train.pred.label!= train.cons.label))
test.cons.miss[r] <- length(which(test.pred.label!= test.cons.label))
valid.cons.miss[r] <- length(which(valid.pred.label!= valid.cons.label))
train.iter.miss[r] <- length(which(train.pred.label!= train.iter.label))
test.iter.miss[r] <- length(which(test.pred.label!= test.iter.label))
valid.iter.miss[r] <- length(which(valid.pred.label!= valid.iter.label))
## Update the label tracker
if(r!=4)
{
label.tracker[miss.iter] <- label.tracker[miss.iter]+1
}
# #cp = complexity parameter
# printcp(train.cl.model) # display the results
# summary(train.cl.model) # detailed summary of splits
# rpart.plot(train.cl.model)
}
total.miss <- rbind(total.iter.miss, total.cons.miss, train.cons.miss,
train.iter.miss, test.cons.miss, test.iter.miss,
valid.cons.miss, valid.iter.miss)
total.miss
#Use same data to generate non-iterative model
#Cross fingers this model will suck
#Variable names are the same as above, this might need to be changed
train.data <- data.frame(cbind(train.iter.label, train.img))
colnames(train.data)[1] <- "label"
train.std.model <- rpart(formula, method = "class", data = train.data)
train.pred.label <- unlist(predict(train.std.model, train.data, type="class"))
test.pred.label <- unlist(predict(train.std.model, test.data, type="class"))
valid.pred.label <- unlist(predict(train.std.model, valid.data, type="class"))
#Store predicted labels
pred.label <- c(train.pred.label, test.pred.label, valid.pred.label)
pred.label <- unlist(pred.label)
##Calculate way too many miss indices (only cons this time)
total.cons.miss.comp <-length(which(pred.label!= cons.used.label))
train.cons.miss.comp <- length(which(train.pred.label!= train.cons.label))
test.cons.miss.comp <- length(which(test.pred.label!= test.cons.label))
valid.cons.miss.comp <- length(which(valid.pred.label!= valid.cons.label))
total.miss.comp <- rbind(total.cons.miss.comp, train.cons.miss.comp,
test.cons.miss.comp, valid.cons.miss.comp)
total.miss.comp