perceptron

type AveragedStrategy struct {
	P, N int64
	// contains filtered or unexported fields
}

type Decoded struct {
	InstanceVal Instance
	DecodedVal  util.Equaler
}

type DecodedInstance interface {
	Instance
	Instance() Instance
	Decoded() interface{}
}

type EarlyUpdateInstanceDecoder interface {
	DecodeEarlyUpdate(i DecodedInstance, m Model) (decoded DecodedInstance, decodedFeatures, goldFeatures interface{}, earlyUpdatedAt, goldSize int, decodeScore float64)
}

type EmptyFeatureExtractor struct {
}

type FeatureExtractor interface {
	Features(instance Instance, flag bool, transType byte, trans_values []int) []Feature
	EstimatedNumberOfFeatures() int
	SetLog(bool)
}

type Instance interface {
	util.Equaler
}

type InstanceDecoder interface {
	Decode(i Instance, m Model) (DecodedInstance, interface{})
	DecodeGold(i DecodedInstance, m Model) (DecodedInstance, interface{})
}

type LinearPerceptron struct {
	Decoder        EarlyUpdateInstanceDecoder
	GoldDecoder    InstanceDecoder
	Updater        UpdateStrategy
	Iterations     int
	Model          Model
	Log            bool
	Tempfile       string
	TrainI, TrainJ int
	TempLines      int

	FailedInstances int

	Continue StopCondition
}

type Model interface {
	// util.Persist
	Score(features interface{}) int64
	Add(features interface{}) Model
	Subtract(features interface{}) Model
	AddSubtract(goldFeatures, decodedFeatures interface{}, amount int64)
	ScalarDivide(int64)
	Copy() Model
	AddModel(Model)
	New() Model
}

type StopCondition func(curIt, numIt, generations int, model Model) bool

type SupervisedTrainer interface {
	Train(instances []DecodedInstance)
}

type TrivialStrategy struct{}

type UnsupervisedTrainer interface {
	Train(instances []Instance)
}

type UpdateStrategy interface {
	Init(m Model, iterations int)
	Update(model Model)
	Finalize(m Model) Model
}