vspatch: was missing the non-reward DA that down-trains NR responses …

…-- this is critical and really impairs learning. now obvious that the main issue is need for opponent D1 / D2 pathways so that D2 learns when not to respond and D1 learns when to respond.

examples/vspatch/vspatch.go

@@ -332,27 +332,24 @@ func (ss *Sim) ApplyPVLV(ev *VSPatchEnv, trial int, di uint32) {
 
 	if trial == ev.NTrials-1 {
 		axon.SetGlbV(ctx, di, axon.GvACh, 1)
-		ss.ApplyRew(ev, di)
+		ss.ApplyRew(di, ev.Rew)
 	} else {
-		axon.GlobalSetRew(ctx, di, 0, false) // no rew
+		ss.ApplyRew(di, 0)
 		axon.SetGlbV(ctx, di, axon.GvACh, 0)
-		axon.SetGlbV(ctx, di, axon.GvDA, 0)
 	}
 }
 
 // ApplyRew applies reward
-func (ss *Sim) ApplyRew(ev *VSPatchEnv, di uint32) {
-	// note: not using RPE here at this point
-	rew := ev.Rew
-	ss.SetRew(rew, di)
-}
-
-func (ss *Sim) SetRew(rew float32, di uint32) {
+func (ss *Sim) ApplyRew(di uint32, rew float32) {
 	ctx := &ss.Context
 	pv := &ss.Net.PVLV
+	if rew != 0 {
+		axon.GlobalSetRew(ctx, di, rew, true)
+	} else {
+		axon.GlobalSetRew(ctx, di, rew, false)
+	}
 	vsp := axon.GlbV(ctx, uint32(di), axon.GvRewPred)
 	dap := rew - vsp
-	axon.GlobalSetRew(ctx, di, rew, true)
 	if rew > 0 {
 		pv.SetUS(ctx, di, axon.Positive, 0, 1)
 	} else if rew < 0 {
@@ -387,7 +384,9 @@ func (ss *Sim) NewRun() {
 func (ss *Sim) InitStats() {
 	ss.Stats.SetFloat("Rew", 0)
 	ss.Stats.SetFloat("RewPred", 0)
-	ss.Stats.SetFloat("RPE", 0)
+	ss.Stats.SetFloat("RewPred_NR", 0)
+	ss.Stats.SetFloat("DA", 0)
+	ss.Stats.SetFloat("DA_NR", 0)
 }
 
 // StatCounters saves current counters to Stats, so they are available for logging etc
@@ -414,19 +413,26 @@ func (ss *Sim) NetViewCounters(tm etime.Times) {
 		ss.TrialStats(di) // get trial stats for current di
 	}
 	ss.StatCounters(di)
-	ss.ViewUpdt.Text = ss.Stats.Print([]string{"Run", "Epoch", "Sequence", "Trial", "Di", "TrialName", "Cycle", "Rew", "RewPred", "RPE"})
+	ss.ViewUpdt.Text = ss.Stats.Print([]string{"Run", "Epoch", "Sequence", "Trial", "Di", "TrialName", "Cycle", "Rew", "RewPred", "RewPred_NR", "DA", "DA_NR"})
 }
 
 // TrialStats records the trial-level statistics
 func (ss *Sim) TrialStats(di int) {
 	ctx := &ss.Context
+	diu := uint32(di)
 	ev := ss.Envs.ByModeDi(ctx.Mode, di).(*VSPatchEnv)
 	ss.Stats.SetInt("Cond", ev.Sequence.Cur)
+	hasRew := (axon.GlbV(ctx, diu, axon.GvHasRew) > 0)
 	ss.Stats.SetFloat32("Rew", ev.Rew)
-	ev.RewPred = axon.GlbV(ctx, uint32(di), axon.GvRewPred)
-	ev.RPE = axon.GlbV(ctx, uint32(di), axon.GvDA)
-	ss.Stats.SetFloat32("RewPred", ev.RewPred)
-	ss.Stats.SetFloat32("RPE", ev.RPE)
+	ev.RewPred = axon.GlbV(ctx, diu, axon.GvRewPred)
+	ev.DA = axon.GlbV(ctx, diu, axon.GvDA)
+	if hasRew {
+		ss.Stats.SetFloat32("RewPred", ev.RewPred)
+		ss.Stats.SetFloat32("DA", ev.DA)
+	} else {
+		ss.Stats.SetFloat32("RewPred_NR", ev.RewPred)
+		ss.Stats.SetFloat32("DA_NR", ev.DA)
+	}
 }
 
 //////////////////////////////////////////////////////////////////////////////
@@ -446,7 +452,13 @@ func (ss *Sim) ConfigLogs() {
 	li.Range.Max = 1.2
 	li = ss.Logs.AddStatAggItem("RewPred", etime.Run, etime.Epoch, etime.Sequence)
 	li.Range.Max = 1.2
-	li = ss.Logs.AddStatAggItem("RPE", etime.Run, etime.Epoch, etime.Sequence)
+	ss.Logs.AddStatAggItem("RewPred_NR", etime.Run, etime.Epoch, etime.Sequence)
+	li = ss.Logs.AddStatAggItem("DA", etime.Run, etime.Epoch, etime.Sequence)
+	li.Range.Min = -0.5
+	li.Range.Max = 1
+	li.FixMin = true
+	li.FixMax = true
+	li = ss.Logs.AddStatAggItem("DA_NR", etime.Run, etime.Epoch, etime.Sequence)
 	li.Range.Min = -0.5
 	li.Range.Max = 1
 	li.FixMin = true
@@ -455,14 +467,15 @@ func (ss *Sim) ConfigLogs() {
 
 	// axon.LogAddDiagnosticItems(&ss.Logs, ss.Net, etime.Epoch, etime.Trial)
 
-	ss.Logs.PlotItems("Rew", "RewPred", "RPE")
+	ss.Logs.PlotItems("Rew", "RewPred", "RewPred_NR", "DA", "DA_NR")
 
 	ss.Logs.CreateTables()
 
 	ss.Logs.SetContext(&ss.Stats, ss.Net)
 	// don't plot certain combinations we don't use
-	// ss.Logs.NoPlot(etime.Train, etime.Cycle)
+	ss.Logs.NoPlot(etime.Train, etime.Cycle)
 	ss.Logs.NoPlot(etime.Train, etime.Trial)
+	ss.Logs.NoPlot(etime.Test, etime.Cycle)
 	ss.Logs.NoPlot(etime.Test, etime.Trial)
 	ss.Logs.NoPlot(etime.Test, etime.Run)
 	// note: Analyze not plotted by default

examples/vspatch/vspatch_env.go

@@ -69,8 +69,8 @@ type VSPatchEnv struct {
 	// reward prediction from model
 	RewPred float32 `edit:"-"`
 
-	// reward prediction error: Rew - RewPred
-	RPE float32 `edit:"-"`
+	// DA = reward prediction error: Rew - RewPred
+	DA float32 `edit:"-"`
 }
 
 func (ev *VSPatchEnv) Name() string {