CODEWITHSUNDEEP
rmatrixmemory
Mark
Mark

Reputation: 1769

How to avoid large matrix allocation

I have two arrays of n elements (arrays p and q):

p1 p2 p3 ...
q1 q2 q3 ...

At first calculate, I should calculate

p1*q1+p2*q2

At second:

p1*q1+p2*q2+p3*q3+p4*q4

At third:

p1*q1+p2*q2+p3*q3+p4*q4+p5*q5+p6*q6

and so on, for k times. These can be view as products between row vector and column vector, so I tried to use the matrix multiplication:

Mat<-t(replicate(k, p))
Mat[col(Mat)>2*row(Mat)] <- 0
Res<-Mat%*%q

Nevertheless, the size of matrix Mat must be too large, because I obtain

Error: cannot allocate vector of size 27.7 Mb

Probably I could solve that problem using bigmemory package, as suggested by several posts on this forum, but I wonder if there is a way to simplify my idea. For example, I observe that with matrix Mat we allocate several entries which we don't use (the 0s).

Upvotes: 1

Views: 205

Answers (1)

Florian
Florian

Reputation: 25435

I think this does what you want:

set.seed(1)
p = sample(1:5,100,replace=T)
q = sample(1:5,100,replace=T)

# my initial idea was the following, but that contains a lot of redundant calculations...
sapply(1:length(p), function(x) {sum(head(q,x)*head(p,x))})

# a lot simpler:
cumsum(p*q)

first element is p[1]*q[1], second element is p[1]*q[1] + p[2]*q[2], etc.

Hope this helps!

Upvotes: 2

Related Questions