This is the author's implementation of the PAMELI algorithm proposed in PAMELI: A Meta-Algorithm for Computationally Expensive Multi-Objective Optimization Problems (pending publication).
Known:
- PyTorch 1.3.0
- Numpy 1.17.5
- Scikit-learn 0.22.1
- Pygmo 2.13.0
- PyDOE 0.3.8
cd to the directory of the repository and run:
python run.py --problem <PROBLEM_NAME>
In the current implementation you can use any of the DTLZ and WFG problems. For example, to test on DTLZ2:
python run.py --problem DTLZ2
PAMELI vs. K-RVEA on the DTLZ problem set (from left to right and from top to bottom: DTLZ1, DTLZ2, DTLZ3, DTLZ4, DTLZ5, DTLZ6 and DTLZ7). The curves show the evolution of the inverted generational distance (IGD) with respect to the number of objective function evaluations:
Evolution for 10 iterations of the approximated Pareto set on the DTLZ2 problem:
