Package edu.uiuc.ge.illigal.gale.config
Configuration of GALE.
See:
Description
Package edu.uiuc.ge.illigal.gale.config Description
Configuration of GALE.
This package contains the objects that contain the configuration information
for GALE runs. The configuration information
is provided in these release through a configuration file. An example of
configuration file is:
#-------------------------
# GALE Configuration file
#-------------------------
###
### Relation: Iris
###
NumAttributes = 4
NumClasses = 3
###
### Information related to the knowledge representation used
###
AgentInitAttributes = 3
Prune = true
TestMode = Best
Distance = Euclidean
Agent = IB
AgentMaxInitWidth = 12
AgentMaxInitDepth = 1
###
### GALE execution parameter
###
# General values
WorldIterations = 150
# Board Mapping
MaxX = 32
MaxY = 32
MappingStrategy = Uniform
# Initializations
Seed = 69
SeedRuns = 69
InitialAggregates = 0.8
# Pameters
MergeP = 0.4
SomaticMutationP = 0.01
kThreshold = -0.25
MaxSplitP = 0.5
These parameters are related to three main issues. Please read
GALE papers if you are not sure of
what they mean.The first one is the domain given to
GALE. The parameters are:
- NumAttributes
The number of attributes of the domain to be solved.
- NumClasses
The number of classes of the domain to be solved.
The second one is related to the knowledge representation used. In this
release only rules, instances, and orthogonal decision trees are available.
- Agent
Choose the knowledge representation to be used. CL stands for rules,
IB for instances, and DT for decision trees.
- AgentInitAttributes
Number of initial attributes used. This value is only used in rules.
- Prune
Boolean that indicates if not used information should be removed.
- TestMode
Should test unseen data using the best individual or bagging? Legal
values are best and bagging.
- Distance
Which distance function should be used? This parameter is only used if
evolving instances. Legal values are: Euclidean, nominal, and mixed.
- AgentMaxInitWidth
This parameter controls some initialization issues. In rules it determines
the number of rules of the random initialized individuals. For instances,
it is the number of random initialized instances of the individuals.
Finally, in an orthogonal decision tree it is not used.
- AgentMaxInitDepth
This parameter controls some initialization issues. In rules it is not
used. Whereas, in instance sets it determines the K of the k-nearest
neighbor algorithm. Finally, in an orthogonal decision tree it bounds
the maximum achievable depth.
Finally, the last set of parameters configure
GALE environment.
- MaxX
Size of the board (x value.)
- MaxY
Size of the board (y value.)
- WorldIterations
Number of iteration to be done.
- MappingStrategy
This parameters control how the instances are going to be spread across
the board. In this distribution only two different strategies are
available: uniform and pyramidal.
- Seed
The seed to be used for cross-validation issues.
- SeedRuns
The seed to be used for the runs.
- InitialAggregates
Proportion of occupied cells in the board.
- MergeP
Merge probability.
- SomaticMutationP
Gene mutation probability.
- kThreshold
The value of the kThreshold in the survival phase.
- MaxSplitP
The maximum splitting probability