This is probably more efficient in high dimensions. I am thinking mostly about 1,2, and 3. I have many routines and want to avoid keeping three versions of each in sync. Your method will perform worse in 1D, and maybe 2D and 3D too. But, I'm starting to think that, for most of my applications, the performance hit may be worth the reduction in code complexity.
On Sunday, September 25, 2016 at 12:09:14 AM UTC+2, David P. Sanders wrote: > > > > El sábado, 24 de septiembre de 2016, 21:26:52 (UTC+2), > lapeyre....@gmail.com escribió: >> >> I want to generate something like an if-elseif-else construct or >> switch-case programmatically. >> >> The use I have in mind is to write an expression for taking a step in a >> random walk in n dimensions, >> so I need 2*n branches. >> > > This is not a good way to solve this problem. Since you are in n > dimensions, just use a vector of length n, > so that x[i] is the position in direction i (rather than generating a > variable called x1, another x2, etc.). Then do something like this: > > julia> n = 10 > > julia> x = zeros(n) > > julia> direction = rand(1:n) > > julia> if rand() < 0.5 > x[direction] += 1 > else > x[direction] -= 1 > end > >> >>
