-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.R
182 lines (173 loc) · 7.15 KB
/
utils.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
########################################################
# Developed by S. Alaimo (alaimos at dmi . unict . it) #
# Released under licence GNU GENERAL PUBLIC LICENSE 3 #
# Date: 2015-06-01 #
########################################################
graphWeights <- function (n, m, A, lambda=0.5) {
if (nrow(A) != n || ncol(A) != m) {
stop("The matrix A should be an n by m matrix.")
}
Ky <- diag(1/colSums(A))
Ky[is.infinite(Ky) | is.na(Ky)] <- 0 #BugFix: 1/0=Infinite replaced with 0
kx <- rowSums(A)
Nx <- 1/(matrix(kx, nrow=n, ncol=n, byrow=TRUE)^(lambda) *
matrix(kx, nrow=n, ncol=n, byrow=FALSE)^(1-lambda))
Nx[is.infinite(Nx) | is.na(Nx)] <- 0 #BugFix: 1/0=Infinite replaced with 0
kx[is.infinite(kx) | is.na(kx)] <- 0 #BugFix: 1/0=Infinite replaced with 0
W <- t(A %*% Ky)
W <- A %*% W
W <- Nx * W
rownames(W) <- rownames(A)
colnames(W) <- rownames(A)
W[is.na(W)] <- 0 #This should never happen
return (W)
}
computeRecommendation <- function (AOT, ATD, lambda1=0.5, lambda2=0.5) {
n <- nrow(AOT)
m <- ncol(AOT)
o <- nrow(ATD)
p <- ncol(ATD)
if (m != o) {
stop(paste0("The number of columns in ncRNA-Target matrix should ",
"match the number of rows in the Target-Disease matrix"))
}
WT <- graphWeights(n=m, m=n, A=t(AOT), lambda=lambda1)
WD <- graphWeights(n=p, m=o, A=t(ATD), lambda=lambda2)
WC <- WT %*% (ATD %*% WD)
R <- AOT %*% WC
return (R)
}
output.table <- function (data) {
library(xtable)
return (HTML(paste(
capture.output(
print(
xtable(data, digits=4),
type = "html",
html.table.attributes="class=\"data table table-bordered table-condensed table-hover\"")),
collapse = "\n")))
}
read.input.file <- function(data, what="") {
file.path <- data[1,"datapath"]
file.type <- tolower(data[1,"type"])
tmp <- NULL
if (file.type == "text/plain") {
tmp <- data.matrix(read.delim(file=file.path, check.names=FALSE))
} else {
obj <- try(readRDS(file.path))
if (is.data.frame(obj)) {
tmp <- data.matrix(obj)
} else if (is.matrix(obj) && all(is.numeric(obj))) {
tmp <- obj
}
}
if (!is.matrix(tmp)) {
stop(paste0(what, "Invalid file format: file does not contain a matrix."))
}
if (nrow(tmp) <= 1 || ncol(tmp) <= 1) {
stop(paste0(what, "Invalid file format: You must load a matrix with more than one row or column."))
}
return (tmp)
}
test.crossvalidation <- function (M1, M2, algorithm, test.runs=1, k.fold=10,
top.number=20, progress=NULL, ...) {
n <- nrow(M1)
m <- ncol(M2)
p <- ncol(M1)
A <- M1 %*% M2
A[A != 0] <- 1
ns <- rownames(A)
ms <- colnames(A)
ps <- colnames(M1)
k.n <- rowSums(A)
k.m <- colSums(A)
valid.n <- unname(which(k.n > 2))
valid.m <- unname(which(k.m > 2))
pairs <- which(A == 1, arr.ind=TRUE, useNames=FALSE)
######
nvalid.pairs <- pairs[which(!(pairs[,1] %in% valid.n)),]
valid.pairs <- pairs[which( pairs[,1] %in% valid.n) ,]
nvalid.pairs <- rbind(nvalid.pairs, valid.pairs[which(!(valid.pairs[,2] %in% valid.m)),])
valid.pairs <- valid.pairs[which(valid.pairs[,2] %in% valid.m),]
n.v.pairs <- nrow(valid.pairs)
######
test.r <- numeric(test.runs*k.fold); test.eP <- numeric(test.runs*k.fold)
test.eR <- numeric(test.runs*k.fold); test.IL <- numeric(test.runs*k.fold)
test.hL <- numeric(test.runs*k.fold)
test.rac <- vector("list", test.runs*k.fold)
real.index <- 1
for (t.i in 1:test.runs) {
folds <- 1:k.fold
id <- sample(folds, n.v.pairs, replace=TRUE)
for (test.fold in folds) {
if (!all(is.null(progress))) {
progress$inc(1/(test.runs*k.fold), detail=paste("Test",t.i,"Fold",test.fold))
}
test.set <- subset(valid.pairs, id %in% test.fold)
train.set <- subset(valid.pairs, id %in% folds[-test.fold])
tmp.M1 <- M1
tmp.M2 <- M2
for (s in 1:nrow(test.set)) {
f <- ns[test.set[s,1]]
e <- ms[test.set[s,2]]
f.e <- names(which(M2[,e] == 1))
tmp.M1[f,f.e][which(M1[f,f.e] == 1)] <- 0
}
new.A <- tmp.M1 %*% tmp.M2
new.A[new.A != 0] <- 1
new.R <- do.call(algorithm, list(tmp.M1, tmp.M2, ...))
rem.e.usrs <- sort(unique(test.set[,2]))
r <- 0; rs <- 0; Pl <- 0; Rl <- 0; TP <- 0; FN <- 0
for (i in rem.e.usrs) {
RR <- sort(new.R[new.A[,i] == 0,i], decreasing=TRUE) #Sort list of i-th user
D <- test.set[test.set[,2] == i, ]; #Get deleted links for i-th user
if (is.matrix(D)) { d.o <- ns[D[,1]]; } else { d.o <- ns[D[1]]; }
p <- which(names(RR) %in% d.o) #recovered links for i-th user
r <- r + sum(p/(n-k.m[i])) #Recovery of deleted links
rs <- rs + sum(RR[1:top.number]/(n-k.m[i])) #Average ranking score
di <- length(p[p <= top.number]) #Mean precision and recall P(top.number) and R(top.number)
Pl <- Pl + (di / top.number) #Mean precision and recall P(top.number) and R(top.number)
Rl <- Rl + (di / length(p)) #Mean precision and recall P(top.number) and R(top.number)
TP <- TP + di #True Positive
FN <- FN + length(p[p > top.number]) #False Positive
}
r <- r / nrow(test.set) #Recovery of deleted links
rs <- rs / length(rem.e.usrs) #Average ranking score
Pl <- Pl / length(rem.e.usrs) #Mean precision and recall P(top.number) and R(top.number)
Rl <- Rl / length(rem.e.usrs) #Mean precision and recall P(top.number) and R(top.number)
eP <- Pl * (n*m/nrow(test.set)) #Precision and recall enhancement, eP(top.number) and eR(top.number)
eR <- Rl * (n/top.number) #Precision and recall enhancement, eP(top.number) and eR(top.number)
###################################################################################################
#Personalization h(top.number)
hL <- 0; c <- 0
for (i in rem.e.usrs) {
RRi <- names(sort(new.R[new.A[,i] == 0,i], decreasing=TRUE)[1:top.number])
for (j in rem.e.usrs) {
if (i != j) {
RRj <- names(sort(new.R[new.A[,j] == 0,j], decreasing=TRUE)[1:top.number])
qij <- length(intersect(RRi, RRj))
hL <- hL + (1-(qij/top.number))
c <- c+1
}
}
}
hL <- hL / c
###################################################################################################
#Surprisal/novelty, I(L)
I <- log2(m/k.n); IL <- 0
I[is.infinite(I)] <- 0
for (i in rem.e.usrs) {
RR <- names(sort(new.R[new.A[,i] == 0,i], decreasing=TRUE)[1:top.number])
IL <- IL + mean(I[RR])
}
IL <- IL / length(rem.e.usrs)
test.r[real.index] <- r ; test.eP[real.index] <- eP
test.eR[real.index] <- eR; test.IL[real.index] <- IL
test.hL[real.index] <- hL;
test.rac[[real.index]] <- new.R
real.index <- real.index + 1
}
}
return (data.frame(r=test.r, eP=test.eP,
eR=test.eR, hL=test.hL, IL=test.IL))
}