|
a |
|
b/predict.voomDDA.R |
|
|
1 |
predict.voomDDA = function(object, newdata){ |
|
|
2 |
|
|
|
3 |
n = ncol(newdata) |
|
|
4 |
p = nrow(newdata) |
|
|
5 |
|
|
|
6 |
disc = matrix(0, n, object$nclass) ## Discriminant scores for each class |
|
|
7 |
dimnames(disc) = list(colnames(newdata), object$classNames) |
|
|
8 |
vm = voomGSD(data.train = object$counts, data.test = newdata, group = object$conditions, norm.method = object$normalization) |
|
|
9 |
x = vm$TestExp |
|
|
10 |
|
|
|
11 |
x2 = t(x) |
|
|
12 |
|
|
|
13 |
{ |
|
|
14 |
if (object$PooledVar) { |
|
|
15 |
vp = (object$weightedStats$weightedSD.pooled)^2 |
|
|
16 |
if (any(i0 <- vp == 0)) |
|
|
17 |
vp[i0] <- 1e-07 * min(vp[!i0]) |
|
|
18 |
ivp <- rep(1/vp, each = n) |
|
|
19 |
for (k in 1:(object$nclass)) { |
|
|
20 |
y = x2 - rep(object$weightedStats$weightedMean.C[, k], each = n) |
|
|
21 |
disc[, k] = rowSums(y * y * ivp) |
|
|
22 |
} |
|
|
23 |
} |
|
|
24 |
|
|
|
25 |
else { |
|
|
26 |
if (FALSE) { |
|
|
27 |
for (k in 1:(object$nclass)) { |
|
|
28 |
x2 = x2 - rep(object$weightedStats$weightedMean.C[, k], each = n) |
|
|
29 |
vsd = (object$weightedStats$weightedSD.C)^2 |
|
|
30 |
disc[, k] = rowSums((x2 * x2)/rep(vsd[, k], each = n)) + sum(log(vsd[, k])) |
|
|
31 |
} |
|
|
32 |
} |
|
|
33 |
else { |
|
|
34 |
vsd = (object$weightedStats$weightedSD.C)^2 |
|
|
35 |
for (k in 1:(object$nclass)) { |
|
|
36 |
disc[, k] = apply(x2, 1, function(z) sum((z - object$weightedStats$weightedMean.C[, k])^2/vsd[, k])) + sum(log(vsd[, k])) |
|
|
37 |
} |
|
|
38 |
} |
|
|
39 |
} |
|
|
40 |
} |
|
|
41 |
|
|
|
42 |
idx = apply(disc, 1, which.min) |
|
|
43 |
pred = colnames(disc)[idx] |
|
|
44 |
if (inherits(attr(x2, "na.action"), "exclude")) |
|
|
45 |
pred = napredict(omit = attr(x2, "na.action"), pred) |
|
|
46 |
pred |
|
|
47 |
} |