Applying exponentiation to vectorized form avoiding for loop

General
1

Hi Everyone,

is there a way to apply exponentiation to vectors without using for loop?

for example,

I have,

model{
vector[100] p;
vector[100] q;
…
q = p^3.02;
…
}

How can I do this without for loop like this,

for (n in 1:100) {
q[n] = p[n] ^ 3.02;
}

Looking forward for an efficient solution to avoid using for loop.

Thanks,

2

There is no elementwise power operator in the Stan language, but if there were it would evaluate to the same loop as

for (n in 1:100) q[n] = p[n] ^ 3.02;
2 Likes
3

so there is no way for me to avoid the for loop when I have to use the power operator?

4

Given that Stan for loops are translated to C++ there is no drop in performance. And I don’t think there is any linear algebra stuff for making AD any faster with elementwise operations (I could be wrong).

Bottom line: for loops are not slow in Stan. In some cases vectorized stuff has better AD properties and are then recommended.

2 Likes
5

I have to use a lot of separate for loops in my model because of the power operator at different parts of my model equation.

Hence, trying to find an alternative to make it vectorized

6

Maybe create a user function for it?

2 Likes