Please enable JavaScript.
Coggle requires JavaScript to display documents.
Parameter Tuning (EA
Parameter (Unordered
Domain (T1) (Categorical,…
Parameter Tuning
EA
Parameter
-
-
-
Hierarchical
Structure
- Have numeric parameter
- Eg: Tournament selection
No Hierarchical
Structure
- No have numeric parameter
- FPS
-
EA vs EA Instances
Different EA Instances:
- Have different symbolic parameter
Variant of same EA
- Have same symbolic parameter
- Different numeric parameter
-
EA Performance
Measure
-
Algorithm Speed
Number of fitness evaluation
CPU time, etc
Variation
- Fix time & measure quality
- Fix quality & measure time
- Fix both & measure completion
-
Parameter Control
p = p(t)
Problem:
- Hard to find optimal p
- Finding optimal p(t) is harder
Vary Mutation
Step Size
-
-
-
Option 3
(x1, …, xn, σ)
σ '= σ * e^(N(0, σ))
x'[i] = x[i] +N(0, σ')
Option 4
(x1, …, xn, σ1, …, σn)
σ'[i] = σ * e^(N(0, i))
x'[i] = x[i] +N(0, σ'[i])
Vary Penalties
eval(x) = f(x) + W × penalty(x)
- 1 for constraint violated
- 0 for constraint satisfied
- W is constant
-
Option 2
W(t+ 1) =
- A x W(t) if last k champions all feasible
- B x W(t) if last k champions all infeasible
- W(t) otherwise
Option 3
(x1, …, xn, W)
eval ((x, W)) = f (x) + W × penalty(x)
Types?
Adaptive
- Feedback rule based on some measure
monitoring search progress
Self-Adaptive
- Parameters evolve along with solutions
- Encoded onto chromosomes they undergo
variation and selection
Deterministic
- rules modified parameter without feedback
Evidence
Absolute Evidence
- Predefined event triggers change
- Direction and magnitude of change is fixed
Relative Evidence
- Compare values through solutions
created with them
- Direction and magnitude of change is not fixed
Influences By
-
Population Statistics
- Mutation successful rate
- Constraint satisfaction
-
Setting values on-line,
during the actual run
-