Documentation
¶
Overview ¶
Package neatns contains Novelty Search implementation for NEAT method of ANN's evolving
Index ¶
- Variables
- type ItemsDistance
- type ItemsDistances
- type NoveltyArchive
- func (a *NoveltyArchive) DumpFittest(w io.Writer) error
- func (a *NoveltyArchive) DumpNoveltyPoints(w io.Writer) error
- func (a *NoveltyArchive) EndOfGeneration()
- func (a *NoveltyArchive) EvaluateIndividualNovelty(org *genetics.Organism, pop *genetics.Population, onlyFitness bool)
- func (a *NoveltyArchive) EvaluatePopulationNovelty(pop *genetics.Population, onlyFitness bool)
- func (a *NoveltyArchive) UpdateFittestWithOrganism(org *genetics.Organism) error
- type NoveltyArchiveOptions
- type NoveltyItem
- type NoveltyItemsByFitness
- type NoveltyMetric
Constants ¶
This section is empty.
Variables ¶
var ( ErrNoNovelItems = errors.New("no novel items to print") ErrNoFittestItems = errors.New("no fittest items to print") )
Functions ¶
This section is empty.
Types ¶
type ItemsDistance ¶
type ItemsDistance struct {
// contains filtered or unexported fields
}
ItemsDistance the structure to hold distance between two items
type ItemsDistances ¶
type ItemsDistances []ItemsDistance
ItemsDistances the sortable list of distances between two items
func (ItemsDistances) Len ¶
func (f ItemsDistances) Len() int
func (ItemsDistances) Less ¶
func (f ItemsDistances) Less(i, j int) bool
func (ItemsDistances) Swap ¶
func (f ItemsDistances) Swap(i, j int)
type NoveltyArchive ¶
type NoveltyArchive struct {
// NovelItems all the novel items we have found so far
NovelItems []*NoveltyItem
// FittestItems all novel items from the fittest organisms found so far
FittestItems NoveltyItemsByFitness
// Generation the current generation
Generation int
// contains filtered or unexported fields
}
NoveltyArchive The novelty archive contains all the novel items we have encountered thus far. Using a novelty metric we can determine how novel a new item is compared to everything currently in the novelty set
func NewNoveltyArchive ¶
func NewNoveltyArchive(threshold float64, metric NoveltyMetric, options NoveltyArchiveOptions) *NoveltyArchive
NewNoveltyArchive creates new instance of novelty archive
func (*NoveltyArchive) DumpFittest ¶
func (a *NoveltyArchive) DumpFittest(w io.Writer) error
DumpFittest dumps collected novelty points of individuals with maximal fitness found during evolution
func (*NoveltyArchive) DumpNoveltyPoints ¶
func (a *NoveltyArchive) DumpNoveltyPoints(w io.Writer) error
DumpNoveltyPoints dumps collected novelty points to the provided writer as JSON
func (*NoveltyArchive) EndOfGeneration ¶
func (a *NoveltyArchive) EndOfGeneration()
EndOfGeneration the steady-state end of generation call
func (*NoveltyArchive) EvaluateIndividualNovelty ¶
func (a *NoveltyArchive) EvaluateIndividualNovelty(org *genetics.Organism, pop *genetics.Population, onlyFitness bool)
EvaluateIndividualNovelty evaluates the novelty of a single individual organism within population and update its fitness (onlyFitness = true) or store individual's novelty item into archive
func (*NoveltyArchive) EvaluatePopulationNovelty ¶
func (a *NoveltyArchive) EvaluatePopulationNovelty(pop *genetics.Population, onlyFitness bool)
EvaluatePopulationNovelty evaluates the novelty of the whole population and update organisms fitness (onlyFitness = true) or store each population individual's novelty items into archive
func (*NoveltyArchive) UpdateFittestWithOrganism ¶
func (a *NoveltyArchive) UpdateFittestWithOrganism(org *genetics.Organism) error
UpdateFittestWithOrganism to maintain list of the fittest organisms so far
type NoveltyArchiveOptions ¶
type NoveltyArchiveOptions struct {
// KNNNoveltyScore how many nearest neighbors to consider for calculating novelty score, i.e., for how many
// neighbors to look at for N-nearest neighbor distance novelty
KNNNoveltyScore int
// FittestAllowedSize the maximal allowed size for fittest items list
FittestAllowedSize int
// ArchiveSeedAmount is the minimal number of seed novelty items to start from
ArchiveSeedAmount int
}
NoveltyArchiveOptions defines options to be used by NoveltyArchive
func DefaultNoveltyArchiveOptions ¶
func DefaultNoveltyArchiveOptions() NoveltyArchiveOptions
DefaultNoveltyArchiveOptions is to create default NoveltyArchiveOptions
type NoveltyItem ¶
type NoveltyItem struct {
// The generation when item was added to archive
Generation int `json:"generation"`
// The ID of associated individual organism */
IndividualID int `json:"individual_id"`
// The fitness of the associated organism
Fitness float64 `json:"fitness"`
// The novelty of this item
Novelty float64 `json:"novelty"`
// The item's age
Age float64 `json:"age"`
// The data associated with item
Data []float64 `json:"data"`
// contains filtered or unexported fields
}
NoveltyItem is the data holder for novel item's genome and phenotype
type NoveltyItemsByFitness ¶
type NoveltyItemsByFitness []*NoveltyItem
NoveltyItemsByFitness the sortable list of novelty items by fitness
func (NoveltyItemsByFitness) Len ¶
func (f NoveltyItemsByFitness) Len() int
func (NoveltyItemsByFitness) Less ¶
func (f NoveltyItemsByFitness) Less(i, j int) bool
func (NoveltyItemsByFitness) Swap ¶
func (f NoveltyItemsByFitness) Swap(i, j int)
type NoveltyMetric ¶
type NoveltyMetric func(x, y *NoveltyItem) float64
NoveltyMetric The novelty metric function type. The function to compare two novelty items and return distance/difference between them
Source Files