Particle swarm fast implementation

See report.pdf for the full description and analysis of our implementation.

This project is based on the method described in the article Particle swarm with radial basis function surrogates for expensive black-box optimization.

Code Formatting

Running git config --local core.hooksPath .githooks will enable the githooks which use Clang Format to cleanup modified files. Please opt-in.

Correctness / Performance

In order to test the code, for both performance and correctness, we have a system set in place to modularly test each component. To build the necessary shared library do the following:

$ cd opus
$ DEBUG=1 PSO_SHARED=1 make

This will make a shared library pso.so (libpso.dylib on MacOS) which you should put somewhere on your LD_LIBRARY (DYLD_LIBRARY_PATH on MacOS) such that Julia can find it. In order to run the suite, do the following:

$ cd /tests
$ julia
julia> ]
(tests) pkg> activate .
<backspace> # this means hit the backspace/delete key to leave Pkg mode
julia> include("src/tests.jl") # this will run the enabled tests, Gavin plans on improving this interface later

Alternatively, a better approach is to run the tests directly through the command line with julia src/tests.jl, which accepts a combination of the following patterns.

julia src/tests PERF
julia src/tests AUTO <path to libs>
julia src/tests TEST

⚠️ the current TEST option is out-of-date. It only tests the LU Solver currently for simplicity as no more GE and TS changes are being made.

Autotuning

To run the autotuning script you must first install Racket and the DSL rash. Then run cd scripts && ./autotune.rkt. This will build shared libraries as needed and then you can test them with julia src/tests AUTO ../lib

🍻

Gavin Gray, Valentin Ogier, Xavier Servot, York Schlabrendorf