-
Notifications
You must be signed in to change notification settings - Fork 2
/
aft_doc.Rmd
235 lines (187 loc) · 7.84 KB
/
aft_doc.Rmd
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
---
title: "Accelerated Failure Time in Xgboost"
author: "Avinash Barnwal"
date: "8/24/2019"
output: html_document
---
```{r setup, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>"
)
options(width=100)
```
This document is about implementing accelerated failure time model under survival modeling in Xgboost.
It supports 3 underlying distributions - **Normal, Logistic and Extreme.**
This works for 3 kind of censored datasets - **Left, Right and Interval censore types.**
We have used the data present here - [Data Repo](https://github.com/avinashbarnwal/GSOC-2019/tree/master/AFT/test/data/neuroblastoma-data-master/data) and
5 datasets are used -
* [ATAC_JV_adipose](https://github.com/avinashbarnwal/GSOC-2019/tree/master/AFT/test/data/neuroblastoma-data-master/data/ATAC_JV_adipose)
* [CTCF_TDH_ENCODE](https://github.com/avinashbarnwal/GSOC-2019/tree/master/AFT/test/data/neuroblastoma-data-master/data/CTCF_TDH_ENCODE)
* [H3K27ac-H3K4me3_TDHAM_BP](https://github.com/avinashbarnwal/GSOC-2019/tree/master/AFT/test/data/neuroblastoma-data-master/data/H3K27ac-H3K4me3_TDHAM_BP)
* [H3K27ac_TDH_some](https://github.com/avinashbarnwal/GSOC-2019/tree/master/AFT/test/data/neuroblastoma-data-master/data/H3K27ac_TDH_some)
* [H3K36me3_AM_immune](https://github.com/avinashbarnwal/GSOC-2019/tree/master/AFT/test/data/neuroblastoma-data-master/data/H3K36me3_AM_immune)
**Source of the dataset** - [Data](https://github.com/tdhock/neuroblastoma-data/tree/master/data/)
Note - Here input labels are **survival times not log of survival times.**
We have tested with different sigma ranging from [1,2,5,10,100] and distribution from normal, logistic and extreme.
```{r}
require(xgboost)
library(ggplot2)
```
##### Functions to Import Dataset
```{r}
data_import <-function(dataname){
filename = paste('https://raw.githubusercontent.com/avinashbarnwal/GSOC-2019/master/AFT/test/data/neuroblastoma-data-master/data/',dataname,'/',sep="")
inputFileName = paste(filename,'inputs.csv',sep="")
labelFileName = paste(filename,'outputs.csv',sep="")
foldsFileName = paste(filename,'cv/equal_labels/folds.csv',sep="")
inputs = read.table(inputFileName,sep=",",header=T,stringsAsFactors = F)
labels = read.table(labelFileName,sep=",",header=T,stringsAsFactors = F)
folds = read.table(foldsFileName,sep=",",header=T,stringsAsFactors = F)
res = list()
res$inputs = inputs
res$labels = labels
res$folds = folds
return(res)
}
```
##### Function for missing values exclusion and converting log of survival times to survival times.
```{r}
data_massage <- function(inputs,labels){
naColumns = colnames(inputs)[colSums(is.na(inputs))>0]
inputs = inputs[ , !(names(inputs) %in% naColumns)]
labels$min.log.lambda = lapply(labels$min.log.lambda,exp)
labels$max.log.lambda = lapply(labels$max.log.lambda,exp)
res = list()
res$inputs = inputs
res$labels = labels
return(res)
}
```
##### Function for getting train and validation datasets.
```{r}
getXY<-function(foldNo,folds,inputs,labels){
test_id = folds[folds$fold==foldNo,'sequenceID']
train_id = folds[folds$fold!=foldNo,'sequenceID']
X = inputs[inputs$sequenceID %in% train_id,]
X = X[,-which(names(X) %in% c("sequenceID"))]
X = as.matrix(X)
X_val = inputs[inputs$sequenceID %in% test_id,]
X_val = X_val[,-which(names(X_val) %in% c("sequenceID"))]
X_val = as.matrix(X_val)
y_label = labels[labels$sequenceID %in% train_id,]
y_label_test = labels[labels$sequenceID %in% test_id,]
y_lower = as.matrix(y_label$min.log.lambda)
y_upper = as.matrix(y_label$max.log.lambda)
y_lower_val = as.matrix(y_label_test$min.log.lambda)
y_upper_val = as.matrix(y_label_test$max.log.lambda)
res = list()
res$X = X
res$X_val = X_val
res$y_lower = y_lower
res$y_lower_val = y_lower_val
res$y_upper = y_upper
res$y_upper_val = y_upper_val
return(res)
}
```
##### Function for getting Parameters.
```{r}
getParam <- function(sigma,distribution,learning_rate){
eval_metric = paste("aft-nloglik@",distribution,",",sigma,sep="")
param = list(learning_rate=learning_rate, aft_noise_distribution=distribution,
nthread = 4, verbosity=0, aft_sigma= sigma,
eval_metric = eval_metric,
objective = "aft:survival")
return(param)
}
```
##### Function for training models.
```{r}
trainModel <- function(foldNo,X,X_val,y_lower,y_lower_val,y_upper,y_upper_val,param,num_round){
dtrain = xgb.DMatrix(X)
setinfo(dtrain,'label_lower_bound', y_lower)
setinfo(dtrain,'label_upper_bound', y_upper)
dtest = xgb.DMatrix(X_val)
setinfo(dtest,'label_lower_bound', y_lower_val)
setinfo(dtest,'label_upper_bound', y_upper_val)
watchlist <- list(eval = dtest, train = dtrain)
bst <- xgb.train(param, dtrain, num_round, watchlist,verbose = 0)
return(bst)
}
```
##### Function for creating plots.
```{r}
createPlot <- function(dataname,data,distribution){
title = paste("Data=",dataname," Distribution=",distribution,sep="")
p = ggplot(data=data,environment = environment()) +
geom_line(aes(x=iter,y=error,colour=type),
data=data,size=1) +
ylab("Error")+xlab("Number of Iteration") + ggtitle(title) + facet_grid(fold~sigma ,scales="free",labeller=label_both)
print(p)
}
```
##### Function for creating data
```{r}
createData <- function(bst,Fold,distribution,sigma){
colnames(bst$evaluation_log) = c('iter','eval','train')
df_eval = data.frame(bst$evaluation_log$iter,bst$evaluation_log$eval)
df_eval$paramter = rep('eval',nrow(df_eval))
df_eval$sigma = rep(sigma,nrow(df_eval))
df_eval$distribution = rep(distribution,nrow(df_eval))
df_eval$fold = rep(Fold,nrow(df_eval))
colnames(df_eval) = c('iter','error','type','sigma','distribution','fold')
df_train = data.frame(bst$evaluation_log$iter,bst$evaluation_log$train)
df_train$paramter = rep('train',nrow(df_train))
df_train$sigma = rep(sigma,nrow(df_train))
df_train$distribution = rep(distribution,nrow(df_train))
df_train$fold = rep(Fold,nrow(df_train))
colnames(df_train) = c('iter','error','type','sigma','distribution','fold')
df = rbind(df_eval,df_train)
return(df)
}
```
##### Function for creating data and plot
```{r}
createDataPlot <- function(dataName,distribution,folds_iter,sigma_range,folds,inputs,labels,result){
for(sigma in sigma_range) {
for(i in folds_iter){
res = getXY(i,folds,inputs,labels)
X = res$X
X_val = res$X_val
y_lower = res$y_lower
y_lower_val = res$y_lower_val
y_upper = res$y_upper
y_upper_val = res$y_upper_val
param = getParam(sigma,distribution,learning_rate)
bst = trainModel(i,X,X_val,y_lower,y_lower_val,y_upper,y_upper_val,param,num_round)
dataIter = createData(bst,i,distribution,sigma)
result = rbind(result,dataIter)
}
}
createPlot(dataName,result,distribution)
}
```
##### Setting Parameters and Result
```{r}
dataNameRange = c('ATAC_JV_adipose','CTCF_TDH_ENCODE','H3K27ac-H3K4me3_TDHAM_BP','H3K27ac_TDH_some','H3K36me3_AM_immune')
# Set Parameters
sigma_range = c(2,5,10,20,50,100)
distribution_range = c('normal','logistic','extreme')
learning_rate = 0.1
num_round = 200
for(dataName in dataNameRange){
res = data_import(dataName)
inputs = res$inputs
labels = res$labels
folds = res$folds
result = data.frame()
resDataMassage = data_massage(inputs,labels)
inputs = resDataMassage$inputs
labels = resDataMassage$labels
folds_iter = unique(folds$fold)
for(distribution in distribution_range){
createDataPlot(dataName,distribution,folds_iter,sigma_range,folds,inputs,labels,result)
}
}
```