terminal_criterion.go

package lr

import (
	"math"
)

/**
 * It's based the paper "Scalable Training of L1-Regularized Log-Linear Models"
 * by Galen Andrew and Jianfeng Gao
 * user: weixuan
 */
type relativeMeanImprCriterion struct {
	minHist     int
	maxHist     int
	tolerance   float64
	improvement float64
	costList    []float64
}

func NewRelativeMeanImprCriterion(tolerance float64) *relativeMeanImprCriterion {
	tc := new(relativeMeanImprCriterion)
	tc.minHist = 5
	tc.maxHist = 10
	tc.costList = make([]float64, 0, tc.maxHist)
	tc.tolerance = tolerance
	return tc
}

func (tc *relativeMeanImprCriterion) calImprovement() float64 {
	sz := len(tc.costList)
	if sz <= tc.minHist {
		return math.MaxFloat32
	}
	first := tc.costList[0]
	last := tc.costList[sz-1]
	impr := (first - last) / float64(sz-1)
	if last != 0 {
		impr = math.Abs(impr / last)
	} else if first != 0 {
		impr = math.Abs(impr / first)
	} else {
		impr = 0
	}
	if sz > tc.maxHist {
		tc.costList = tc.costList[1:]
	}
	return impr
}

func (tc *relativeMeanImprCriterion) addCost(latestCost float64) {
	tc.costList = append(tc.costList, latestCost)
	tc.improvement = tc.calImprovement()
}

func (tc *relativeMeanImprCriterion) isTerminable() bool {
	return tc.improvement <= tc.tolerance
}