func (this *Environment) EnvStep(action rlglue.Action) (obs rlglue.Observation, r float64, t bool) {
fps := make([]float64, len(this.status))
reboot := int(action.Ints()[0])
for i := range this.status {
if reboot == i {
fps[i] = 0
} else {
fps[i] = this.cfg.FailBase
li := (i + this.cfg.NumSystems - 1) % this.cfg.NumSystems
ri := (i + 1) % this.cfg.NumSystems
if !this.status[li] {
fps[i] += this.cfg.FailIncr
}
if !this.status[ri] {
fps[i] += this.cfg.FailIncr
}
}
if this.status[i] || reboot == i {
this.status[i] = stat.NextUniform() < (1 - fps[i])
} else {
this.status[i] = stat.NextUniform() < (1 - this.cfg.FailStay)
}
if this.status[i] {
r++
}
}
if reboot < this.cfg.NumSystems {
r--
}
obs = this.ConstructObs()
return
}
func (this *Env) EnvStep(action rlglue.Action) (obs rlglue.Observation, r float64, t bool) {
obs = this.obs
var o discrete.Oracle
a := discrete.Action(action.Ints()[0])
o, r = this.belief.Next(a)
this.belief = o.(*Belief)
t = false
return
}
func (ge *Env) EnvStep(action rlglue.Action) (obs rlglue.Observation, r float64, t bool) {
t = ge.Pos.Equals(ge.Goal)
r = -1
if t {
r = 0
}
dir := action.Ints()[0]
u := rand.Float64()
if u < .1 {
dir += 1
} else if u < .2 {
dir += 3
}
dir %= 4
newPos := ge.Pos
switch dir {
case 0:
newPos.Y++
case 1:
newPos.X++
case 2:
newPos.Y--
case 3:
newPos.X--
}
if newPos.X < 0 {
newPos.X = 0
}
if newPos.Y < 0 {
newPos.Y = 0
}
if newPos.X >= ge.Width {
newPos.X = ge.Width - 1
}
if newPos.Y >= ge.Height {
newPos.Y = ge.Height - 1
}
ge.Pos = newPos
obs = rlglue.NewObservation([]int32{ge.Pos.X, ge.Pos.Y}, []float64{}, []byte{})
return
}
func (this *Env) EnvStep(action rlglue.Action) (obs rlglue.Observation, r float64, t bool) {
lastObs := this.MakeObs()
println(action.Ints()[0])
switch action.Ints()[0] {
case 0:
this.hunter, r, t = this.hunter.TurnLeft()
case 1:
this.hunter, r, t = this.hunter.TurnRight()
case 2:
this.hunter, r, t = this.hunter.Move()
case 3:
this.hunter, r, t = this.hunter.Shoot()
}
if !t {
obs = this.MakeObs()
fmt.Fprintf(os.Stderr, "Sending back\n%v\n%v\n", this.hunter, this.observed)
} else {
obs = lastObs
}
return
}
func (this *BFS3Agent) getActionIndex(act rlglue.Action) (index uint64) {
return this.task.Act.Ints.Index(act.Ints())
}
func (this *Env) EnvStep(action rlglue.Action) (obs rlglue.Observation, r float64, t bool) {
whichCan := action.Ints()[0]
process := action.Ints()[1]
t = true
for _, can := range this.Cans {
if !can.Done {
t = false
}
}
this.Log("%v ", this.Cans)
if t {
this.Log("finished\n\n")
obs = this.makeObs()
r = 0
return
}
can := this.Cans[whichCan]
r = -1
if !can.Done {
switch process {
case 0:
this.Log("painting can %d\n", whichCan+1)
outcome := stat.NextChoice([]float64{.6, .3, .1})
switch outcome {
case 0:
can.Painted = true
case 1:
can.Painted = true
can.Scratched = true
case 2:
}
case 1:
this.Log("polishing can %d\n", whichCan+1)
outcome := stat.NextChoice([]float64{.2, .2, .3, .2, .1})
switch outcome {
case 0:
can.Painted = false
case 1:
can.Scratched = false
case 2:
can.Polished = true
can.Painted = false
can.Scratched = false
case 3:
can.Polished = true
can.Painted = false
case 4:
}
case 2:
this.Log("shortcut can %d\n", whichCan+1)
outcome := stat.NextChoice([]float64{0.05, 0.95})
switch outcome {
case 0:
can.Painted = true
can.Polished = true
case 1:
}
case 3:
this.Log("finishing can %d\n", whichCan+1)
if can.Painted && can.Polished && !can.Scratched && !can.Done {
can.Done = true
r = 10
} else {
t = true
r = -100000
}
}
this.Cans[whichCan] = can
}
obs = this.makeObs()
return
}