NSGA-II Algorithm

Forked from wreszelewski/nsga2 but mostly rewritten.

Usage

cwd: Repository base.

Windows:

set PYTHONPATH=%PYTHONPATH%;%cd% && python examples

Unix:

$ export PYTHONPATH=${PYTHONPATH};$(pwd) && python examples

Changelist

The following changes have been made to the pre-fork repository:

1. Compatible with > 3 objective problems.
   • The original nsga2 repository was only compatible with 2-objective problems.
2. Does not require the user to rescale
   • The original nsga2 implementation relied on the decision vectors to be [0,1] but this implementation can take customized bounds.
   • The previous implementation also required the objective space vectors to be standardized but this implementation abstracts over that.
3. The visualizer can now visualize up to 3 dimensions(3 objectives)
4. Removed the use of callback functions and used generators instead.
5. NSGA2 now inherits from an ioptimizer interface which is compatible with iproblem interfaces
   • This was in order to test multiple optimizers in a generic fashion.
6. Python3 Compatible
   • Certain uses of sorted(cmp=) made the original repository incompatible with python3 but it now runs on python3
   • Also, there is occasional use of type hinting especially around the public API

To Do

• Write test code, since debugging this was actually quite difficult.
• Remove the use of state and try to make the methods static methods.
• Matplotlib hangs