func (this *DepLearner) Consider(np ParentSet) {
if np == this.parents {
return
}
if stat.NextBernoulli(1/(1+float64(this.consistency))) == 0 {
return
}
alternateRanges := np.CutRanges(this.bg.ranges)
alternateHistory := this.MakeMappedHistory(np, alternateRanges)
alternateLoglihood := this.MappedLoglihoodRatio(alternateHistory)
choiceLL := []float64{this.mappedLoglihood, alternateLoglihood}
sizeDiff := float64(this.parents.Size(uint32(len(this.bg.ranges))) - np.Size(uint32(len(this.bg.ranges))))
if sizeDiff > 0 {
choiceLL[0] += sizeDiff * this.bg.cfg.Kappa
choiceLL[1] += sizeDiff * (1 - this.bg.cfg.Kappa)
} else {
choiceLL[0] -= sizeDiff * (1 - this.bg.cfg.Kappa)
choiceLL[1] -= sizeDiff * this.bg.cfg.Kappa
}
if stat.NextLogChoice(choiceLL) == 1 {
this.parents = np
this.cutRanges = alternateRanges
this.mappedHistory = alternateHistory
this.mappedLoglihood = alternateLoglihood
this.consistency = 0
} else {
this.consistency++
}
}
func (this *Environment) EnvInit() (taskString string) {
fstr := "VERSION RL-Glue-3.0 PROBLEMTYPE episodic DISCOUNTFACTOR %f OBSERVATIONS INTS (%d 0 1) ACTIONS INTS (0 %d) REWARDS (-1.0 1.0)"
taskString = fmt.Sprintf(fstr, this.cfg.DiscountFactor, this.cfg.NumSystems, this.cfg.NumSystems)
this.task, _ = rlglue.ParseTaskSpec(taskString)
this.status = make([]bool, this.cfg.NumSystems)
for i := range this.status {
this.status[i] = stat.NextBernoulli(this.cfg.StartBoot) == 1
}
return
}