func (this *ROARAgent) AgentStep(reward float64, obs rlglue.Observation) rlglue.Action {
last := matrix.MakeDenseMatrix(this.LastObs.Doubles(), this.numFeatures, 1)
current := matrix.MakeDenseMatrix(obs.Doubles(), this.numFeatures, 1)
rm := matrix.MakeDenseMatrix([]float64{reward}, 1, 1)
outcome, _ := current.MinusDense(last)
sor, _ := last.Augment(outcome)
sor, _ = sor.Augment(rm)
actionIndex := this.task.Act.Ints.Index(this.LastAct.Ints())
this.rpost[actionIndex].Insert(sor)
this.LastObs = obs
return this.GetAction()
}
