Alexey Stukalov

@matthieugomez I think we should be able to safely combine preallocation and parallelization -- we just need to be sure that each worker receives its own copy of "functional object"...

Definitely that would be very nice to have. But that requires more advanced communication to solve the load balancing problems, because the current scheme assumes migration rates are the same...

@steven-varga Hi, if you want to distribute the calculations over several machines, MPI could be a way to go (you can also consider ZeroMQ or use [non-local cluster managers](http://julia.readthedocs.org/en/latest/manual/parallel-computing/?highlight=parallel#clustermanagers) --...

I've updated the code to use `Channel`s very recently introduced into Julia 0.4-dev, so `MessageUtils` are no longer required. However, the "smoketest" using parallel optimizer stalls for some yet-to-be-investigated reason.

Updated to use recently introduced type-parameterized RemoteRefs and parallel exception handling. The tests now pass on 0.4-nightly (the current CI failure is due to XNES smoketest randomly failed). It starts...

@robertfeldt ParallelEvaluator covers methods that generate multiple candidates at once, e.g. NES. This master/slave optimizer [better] fits DE-like algorithms. So both can co-exist. I had not rebased the PR recently...

> How did you reproduce this in VS Code? These were the required checkboxes. I think local VS Code doesn't apply here, and I don't have access to GitHub Codespaces....

doesn't work for me on ArchLinux+RStudio either (latest R & rstan). it would be nice to fix it, because otherwise debugging transformed data generation becomes very inconvenient

It's fine to allow `dims`, especially if the implementation can choose the optimal algorithm depending on the data layout. But many methods are only optimized for the "observations as columns"...

`hclust_minimum()` method really uses the `d` in the "dense" way updating all elements of one row/col per iteration, so I'm not sure that * `SparseMatrixCSC` is the right type for...