I've been trying to modify bradley's code to do something much like that,
but I keep running into an error. I can get it to work on one core, but I
am having trouble getting pmap to work using Bradley's example. Basically
rather than parallelizing a single process that requires the same data over
and over. I want to parallelize a process where each core requires the same
data X AND each core gets a different unique data vector.
I can get it to work just fine on one core (in the example below that means
samples_pmap_core_1 is correct), but I soon as I try Brad's method with
pmap. It doesnt seem to work (i.e. samples_pmap has an error that the
function is not defined on the worker). I feel like a solution here would
help with the OP's as well.
Best,
Alex
Here is the separate file that is required on each core. Called
"needed_all_over.jl"
## Data that stays the same
X=rand(100,1)
##Function
function square(x)
estimate = mean(x.^2)
return estimate
end
## Estimate Basic Function Here
sim_estimates=zeros(sim_reps_node,1);
function square_with_rand(error)
for i=1:sim_reps_node
sim_estimates[i,1]=square([X ; error[i]])
end
return sim_estimates
end
Here is the file that runs on that outside.
srand(2);
## Generate Data that is different across draws (Error Sequence)
## The number of Draws
sim_reps_main=100;
sim_reps_node=int(sim_reps_main/2);
## Generate the std normal error sequence
error=rand(sim_reps_main);
require("needed_all_over.jl")
samples_pmap_core_1 = vcat(square_with_rand(error[1:50]),
square_with_rand(error[1:50]))
addprocs(1)
nprocs() # Now I am doing this on two cores.
samples_pmap = pmap(square_with_rand,[error[1:50],error[51:100]])
samples =
vcat(samples_pmap[1],samples_pmap[2],samples_pmap[3],samples_pmap[4])
