Continue from the last week
from GA
First of all we begin with Island GA
Island GA is an evolving multiple-population in tandem, a form of parallel GA.
At a fix number of generation, some individual from each island
GA for Multi-Objective problems
-Nondominated sorting Genetic Algorithm(NSGA)
-Fast Non-dominated Sorting Genetic Algorithm(NSAGA-II)
-Strength Pareto Evolutionary Stragetegy
GA on CUDA - Nvidia
Holland's Schema Theorem
EC in Java
So this is the End of GA so next is GP (normally it's like adapted from GA by using graph)
GP : Genetic Programming
-Santa Fe Trail
-GP for Tetris
-Alignment
-Modal Structure
-Classification
Mean Square Error (MSE) = 1/N Ej=1to N (targetj-outputj)^2
Mean Absolute Percentage Error (MAPE) =1/N Ej=1toN |(targetj-outputj)/xj|
POCID =100/N Ej=1toN Dj,
where dj{1,if(targetj-outputj-1)(outputj-outputj-1)>0
0, otherwise}
Parse Treeinternal node is an operator +-*/
leave node is an value
Each program is represented as a tree
Population
- Grow - limit and Full - teminal (complete tree)
Crossover in GP
- Randomly select two trees A and B from the population
- A branch is selected randomly in each tree to be cut
- The sub-tree of both A and B are exchanged.
Mutation in GP
- Randomized Generated sub-tree
- Functional node mutation
- Treminal
- Shrink Mutation - randomly selected subtree into new era
- Hoist Mutation
- Survivor selection
End of GP
ไม่มีความคิดเห็น:
แสดงความคิดเห็น