Zhe's twist to MOEAD - Githubissues

vivekaxl / MOLearner

Multi-objective learning for configurations

1 stars 0 forks source link

Zhe's twist to MOEAD #27

Open vivekaxl opened 7 years ago

vivekaxl commented 7 years ago

Algorithm:

generate 20 points
generate 10 random weights (a1x1 + b1x2) -> a = <10>, b = <10>
Policy 1: 1 cart for each direction
Policy 2: 1 cart for each objective (objective < direction)
at each iteration, sample 10 points rather than 1 point. one of each direction.

Advantages: No use of NDS aka fast

Experiments:

Directions is directly proportional to the diversity (directions and diversity are related) - 2, 3, 5, 8, 10 (N+1, N is the number of objectives)

Keep an eye for duplicate points -> directions can be merged?

vivekaxl commented 7 years ago

@azhe825