func initChainFB(t *testing.T) {
var err error
h0 := narray.New(nstates[0], nstates[0])
h1 := narray.New(nstates[1], nstates[1])
h0.Set(1, 0, 1)
h0.Set(.5, 1, 1)
h0.Set(.5, 1, 2)
h0.Set(.3, 2, 2)
h0.Set(.6, 2, 3)
h0.Set(.1, 2, 4)
h0.Set(.7, 3, 3)
h0.Set(.3, 3, 4)
h1.Set(1, 0, 1)
h1.Set(.3, 1, 1)
h1.Set(.2, 1, 2)
h1.Set(.5, 1, 3)
h1.Set(.6, 2, 2)
h1.Set(.4, 2, 3)
h0 = narray.Log(nil, h0.Copy())
h1 = narray.Log(nil, h1.Copy())
ms, _ = NewSet()
hmm0, err = ms.NewNet("model 0", h0,
[]model.Modeler{nil, newScorer(0, 1), newScorer(0, 2), newScorer(0, 3), nil})
fatalIf(t, err)
hmm1, err = ms.NewNet("model 1", h1,
[]model.Modeler{nil, newScorer(1, 1), newScorer(1, 2), nil})
fatalIf(t, err)
}
func makeHMM(t *testing.T) *Model {
// Gaussian 1.
mean1 := []float64{1}
sd1 := []float64{1}
g1 := gm.NewModel(1, gm.Name("g1"), gm.Mean(mean1), gm.StdDev(sd1))
// Gaussian 2.
mean2 := []float64{4}
sd2 := []float64{2}
g2 := gm.NewModel(1, gm.Name("g2"), gm.Mean(mean2), gm.StdDev(sd2))
var err error
h0 := narray.New(4, 4)
// h0.Set(.8, 0, 1)
h0.Set(1, 0, 1)
// h0.Set(.2, 0, 2)
h0.Set(.5, 1, 1)
h0.Set(.5, 1, 2)
h0.Set(.7, 2, 2)
h0.Set(.3, 2, 3)
h0 = narray.Log(nil, h0.Copy())
ms, _ = NewSet()
_, err = ms.NewNet("hmm0", h0,
[]model.Modeler{nil, g1, g2, nil})
fatalIf(t, err)
return NewModel(OSet(ms), UpdateTP(true), UpdateOP(true))
}
// makes left-to-right hmm with self loops.
// ns is the total number of states including entry/exit.
// selfProb is the prob of the self loop with value between 0 and 1.
// skipProb is the prob of skipping next state. Make it zero for no skips.
// for ns=3, skipProb is set to zero.
func (ms *Set) makeLeftToRight(name string, ns int, selfProb,
skipProb float64, dists []model.Modeler) (*Net, error) {
if selfProb >= 1 || skipProb >= 1 || selfProb < 0 || skipProb < 0 {
panic("probabilities must have value >= 0 and < 1")
}
if selfProb+skipProb >= 1 {
panic("selfProb + skipProb must be less than 1")
}
if len(dists) != ns {
panic("length of dists must match number of states")
}
if ns < 3 {
panic("min number of states is 3")
}
p := selfProb
var q float64
if ns > 3 {
q = skipProb
} else {
q = 0
}
r := 1.0 - p - q
h := narray.New(ns, ns)
// state 0
h.Set(1-q, 0, 1) // entry
h.Set(q, 0, 2) // skip first emmiting state
// states 1..ns-3
for i := 1; i < ns-2; i++ {
h.Set(p, i, i) // self loop
h.Set(r, i, i+1) // to right
h.Set(q, i, i+2) // skip
}
// state ns-2
h.Set(p, ns-2, ns-2) // self
h.Set(1-p, ns-2, ns-1) // to exit (no skip)
// convert to log.
h = narray.Log(nil, h.Copy())
hmm, err := ms.NewNet(name, h, dists)
if err != nil {
return nil, err
}
return hmm, nil
}
// randTrans generates a left-to right random transition prob matrix.
// n is the total number of states including entry/exit.
func randTrans(r *rand.Rand, n int, skip bool) *narray.NArray {
if n < 3 {
panic("need at least 3 states")
}
a := narray.New(n, n)
// state 0
if n > 3 && skip {
p := getProbs(r, 2)
a.Set(p[0], 0, 1) // entry
a.Set(p[1], 0, 2) // skip first emmiting state
} else {
a.Set(1, 0, 1) // entry
}
// states 1..n-3
for i := 1; i < n-2; i++ {
if skip {
p := getProbs(r, 3)
a.Set(p[0], i, i) // self loop
a.Set(p[1], i, i+1) // to right
a.Set(p[2], i, i+2) // skip
} else {
p := getProbs(r, 2)
a.Set(p[0], i, i) // self loop
a.Set(p[1], i, i+1) // to right
}
}
// state ns-2
p := getProbs(r, 2)
a.Set(p[0], n-2, n-2) // self
a.Set(p[1], n-2, n-1) // to exit (no skip)
if glog.V(8) {
st := a.Sprint(func(na *narray.NArray, k int) bool {
if na.Data[k] > 0 {
return true
}
return false
})
glog.Info(st)
}
return narray.Log(nil, a)
}
// MakeLeftToRight creates a transition probability matrix for a left-to-right HMM.
// * ns is the total number of states including entry/exit. Must be 3 or greater.
// * selfProb is the prob of the self loop with value between 0 and 1.
// * skipProb is the prob of skipping next state. Make it zero for no skips.
// * for ns=3, skipProb is set to zero.
func MakeLeftToRight(ns int, selfProb, skipProb float64) *narray.NArray {
if selfProb >= 1 || skipProb >= 1 || selfProb < 0 || skipProb < 0 {
panic("probabilities must have value >= 0 and < 1")
}
if selfProb+skipProb >= 1 {
panic("selfProb + skipProb must be less than 1")
}
if ns < 3 {
panic("min number of states is 3")
}
p := selfProb
var q float64
if ns > 3 {
q = skipProb
} else {
q = 0
}
r := 1.0 - p - q
h := narray.New(ns, ns)
// state 0
h.Set(1-q, 0, 1) // entry
h.Set(q, 0, 2) // skip first emmiting state
// states 1..ns-3
for i := 1; i < ns-2; i++ {
h.Set(p, i, i) // self loop
h.Set(r, i, i+1) // to right
h.Set(q, i, i+2) // skip
}
// state ns-2
h.Set(p, ns-2, ns-2) // self
h.Set(1-p, ns-2, ns-1) // to exit (no skip)
return narray.Log(h, h)
}
// should be equivalent to training a single gaussian, great for debugging.
func TestSingleState(t *testing.T) {
// HMM to generate data.
g01 := gm.NewModel(1, gm.Name("g01"), gm.Mean([]float64{0}), gm.StdDev([]float64{1}))
h0 := narray.New(3, 3)
h0.Set(1, 0, 1)
h0.Set(.8, 1, 1)
h0.Set(.2, 1, 2)
h0 = narray.Log(nil, h0.Copy())
ms0, _ := NewSet()
net0, e0 := ms0.NewNet("hmm", h0,
[]model.Modeler{nil, g01, nil})
fatalIf(t, e0)
hmm0 := NewModel(OSet(ms0))
_ = hmm0
// Create gaussian to estimate without using the HMM code.
g := gm.NewModel(1, gm.Name("g1"), gm.Mean([]float64{-1}), gm.StdDev([]float64{2}))
// Create initial HMM and estimate params from generated data.
g1 := gm.NewModel(1, gm.Name("g1"), gm.Mean([]float64{-1}), gm.StdDev([]float64{2}))
h := narray.New(3, 3)
h.Set(1, 0, 1)
h.Set(.5, 1, 1)
h.Set(.5, 1, 2)
h = narray.Log(nil, h.Copy())
ms, _ = NewSet()
net, e := ms.NewNet("hmm", h,
[]model.Modeler{nil, g1, nil})
fatalIf(t, e)
hmm := NewModel(OSet(ms), UpdateTP(true), UpdateOP(true))
iter := 5
// number of sequences
m := 1000
numFrames := 0
t0 := time.Now() // Start timer.
for i := 0; i < iter; i++ {
t.Logf("iter [%d]", i)
// Make sure we generate the same data in each iteration.
r := rand.New(rand.NewSource(33))
gen := newGenerator(r, false, net0)
// Reset all counters.
hmm.Clear()
g.Clear()
// fix the seed to get the same sequence
for j := 0; j < m; j++ {
obs, states := gen.next("oid-" + fi(j))
numFrames += len(states) - 2
hmm.UpdateOne(obs, 1.0)
// Update Gaussian
for _, o := range obs.ValueAsSlice() {
vec := o.([]float64)
gobs := model.NewFloatObs(vec, model.SimpleLabel(""))
g.UpdateOne(gobs, 1.0)
}
}
hmm.Estimate()
g.Estimate()
t.Logf("iter:%d, hmm g1: %+v", i, net.B[1])
t.Logf("iter:%d, direct g1:%+v", i, g)
}
dur := time.Now().Sub(t0)
tp0 := narray.Exp(nil, h0.Copy())
tp := narray.Exp(nil, net.A.Copy())
ns := tp.Shape[0]
for i := 0; i < ns; i++ {
for j := 0; j < ns; j++ {
p0 := tp0.At(i, j)
logp0 := h0.At(i, j)
p := tp.At(i, j)
logp := h.At(i, j)
if p > smallNumber || p0 > smallNumber {
t.Logf("TP: %d=>%d, p0:%5.2f, p:%5.2f, logp0:%8.5f, logp:%8.5f", i, j, p0, p, logp0, logp)
}
}
}
t.Log("")
t.Logf("hmm0 g1:%+v", net0.B[1])
t.Logf("hmm g1: %+v", net.B[1])
t.Log("")
t.Logf("direct g1:%+v", g)
// Print time stats.
t.Log("")
t.Logf("Total time: %v", dur)
t.Logf("Time per iteration: %v", dur/time.Duration(iter))
t.Logf("Time per frame: %v", dur/time.Duration(iter*numFrames*m))
gjoa.CompareSliceFloat(t, tp0.Data, tp.Data,
"error in Trans Probs [0]", .03)
CompareGaussians(t, net0.B[1].(*gm.Model), net.B[1].(*gm.Model), 0.03)
if t.Failed() {
t.FailNow()
}
// Recognize.
sg := ms.SearchGraph()
dec, e := graph.NewDecoder(sg)
if e != nil {
t.Fatal(e)
}
r := rand.New(rand.NewSource(5151))
gen := newGenerator(r, true, net0)
// testDecoder(t, gen, dec, 1000)
testDecoder(t, gen, dec, 10)
}
func TestHMMGauss(t *testing.T) {
// Create reference HMM to generate observations.
g01 := gm.NewModel(1, gm.Name("g01"), gm.Mean([]float64{0}), gm.StdDev([]float64{1}))
g02 := gm.NewModel(1, gm.Name("g02"), gm.Mean([]float64{16}), gm.StdDev([]float64{2}))
h0 := narray.New(4, 4)
h0.Set(.6, 0, 1)
h0.Set(.4, 0, 2)
h0.Set(.9, 1, 1)
h0.Set(.1, 1, 2)
h0.Set(.7, 2, 2)
h0.Set(.3, 2, 3)
h0 = narray.Log(nil, h0.Copy())
ms0, _ := NewSet()
net0, e0 := ms0.NewNet("hmm", h0,
[]model.Modeler{nil, g01, g02, nil})
fatalIf(t, e0)
hmm0 := NewModel(OSet(ms0), UpdateTP(true), UpdateOP(true))
_ = hmm0
// Create random HMM and estimate params from obs.
g1 := gm.NewModel(1, gm.Name("g1"), gm.Mean([]float64{-1}), gm.StdDev([]float64{2}))
g2 := gm.NewModel(1, gm.Name("g2"), gm.Mean([]float64{18}), gm.StdDev([]float64{4}))
h := narray.New(4, 4)
h.Set(.5, 0, 1)
h.Set(.5, 0, 2)
h.Set(.5, 1, 1)
h.Set(.5, 1, 2)
h.Set(.5, 2, 2)
h.Set(.5, 2, 3)
h = narray.Log(nil, h.Copy())
ms, _ = NewSet()
net, e := ms.NewNet("hmm", h,
[]model.Modeler{nil, g1, g2, nil})
fatalIf(t, e)
hmm := NewModel(OSet(ms), UpdateTP(true), UpdateOP(true))
iter := 10
// number of sequences
m := 500
numFrames := 0
t0 := time.Now() // Start timer.
for i := 0; i < iter; i++ {
t.Logf("iter [%d]", i)
// Make sure we generate the same data in each iteration.
r := rand.New(rand.NewSource(33))
gen := newGenerator(r, false, net0)
// Reset all counters.
hmm.Clear()
// fix the seed to get the same sequence
for j := 0; j < m; j++ {
obs, states := gen.next("oid-" + fi(j))
numFrames += len(states) - 2
hmm.UpdateOne(obs, 1.0)
}
hmm.Estimate()
}
dur := time.Now().Sub(t0)
tp0 := narray.Exp(nil, h0.Copy())
tp := narray.Exp(nil, net.A.Copy())
ns := tp.Shape[0]
for i := 0; i < ns; i++ {
for j := 0; j < ns; j++ {
p0 := tp0.At(i, j)
logp0 := h0.At(i, j)
p := tp.At(i, j)
logp := h.At(i, j)
if p > smallNumber || p0 > smallNumber {
t.Logf("TP: %d=>%d, p0:%5.2f, p:%5.2f, logp0:%8.5f, logp:%8.5f", i, j, p0, p, logp0, logp)
}
}
}
t.Log("")
t.Logf("hmm0 g1:%+v, g2:%+v", net0.B[1], net0.B[2])
t.Logf("hmm g1: %+v, g2:%+v", net.B[1], net.B[2])
// Print time stats.
t.Log("")
t.Logf("Total time: %v", dur)
t.Logf("Time per iteration: %v", dur/time.Duration(iter))
t.Logf("Time per frame: %v", dur/time.Duration(iter*numFrames*m))
gjoa.CompareSliceFloat(t, tp0.Data, tp.Data,
"error in Trans Probs [0]", .03)
CompareGaussians(t, net0.B[1].(*gm.Model), net.B[1].(*gm.Model), 0.03)
CompareGaussians(t, net0.B[2].(*gm.Model), net.B[2].(*gm.Model), 0.03)
if t.Failed() {
t.FailNow()
}
// Recognize.
g := ms.SearchGraph()
dec, e := graph.NewDecoder(g)
if e != nil {
t.Fatal(e)
}
r := rand.New(rand.NewSource(5151))
gen := newGenerator(r, true, net0)
testDecoder(t, gen, dec, 1000)
}
func TestMain(m *testing.M) {
// Configure glog. Example to set debug level 6 for file viterbi.go and 3 for everythign else:
// export GLOG_LEVEL=3
// go test -v -run TestTrainHmmGau -vmodule=viterbi=6 > /tmp/zzz
flag.Set("alsologtostderr", "true")
flag.Set("log_dir", "/tmp/log")
level := os.Getenv("GLOG_LEVEL")
if len(level) == 0 {
level = "0"
}
flag.Set("v", level)
glog.Info("glog debug level is: ", level)
ns = 5 // max num states in a model
nstates[0] = 5
nstates[1] = 4
nobs = len(obs)
a = narray.New(nq, ns, ns)
b = narray.New(nq, ns, nobs)
alpha = narray.New(nq, ns, nobs)
beta = narray.New(nq, ns, nobs)
a.Set(1, 0, 0, 1)
a.Set(.5, 0, 1, 1)
a.Set(.5, 0, 1, 2)
a.Set(.3, 0, 2, 2)
a.Set(.6, 0, 2, 3)
a.Set(.1, 0, 2, 4)
a.Set(.7, 0, 3, 3)
a.Set(.3, 0, 3, 4)
a.Set(1, 1, 0, 1)
a.Set(.3, 1, 1, 1)
a.Set(.2, 1, 1, 2)
a.Set(.5, 1, 1, 3)
a.Set(.6, 1, 2, 2)
a.Set(.4, 1, 2, 3)
dist := [][]float64{{.4, .5, .1}, {.3, .5, .2}} // prob dist for states in model 0 and 1
// output probs as a function of model,state,time
for q := 0; q < nq; q++ {
for i := 1; i < nstates[q]-1; i++ {
for t := 0; t < nobs; t++ {
p := dist[q][obs[t]]
// k := r.Intn(len(someProbs))
// b.Set(someProbs[k], q, i, t)
b.Set(p, q, i, t)
}
}
}
// same output probs but as a function of model,state,symbol
// we need this to test the network implementation.
outputProbs = narray.New(nq, ns, nsymb)
for q := 0; q < nq; q++ {
for i := 1; i < nstates[q]-1; i++ {
for k := 0; k < nsymb; k++ {
p := math.Log(dist[q][k])
outputProbs.Set(p, q, i, k)
}
}
}
loga = narray.Log(loga, a.Copy())
logb = narray.Log(logb, b.Copy())
data := make([][]float64, len(obs), len(obs))
for k, v := range obs {
data[k] = []float64{float64(v)}
}
xobs = model.NewFloatObsSequence(data, model.SimpleLabel(""), "")
os.Exit(m.Run())
}
func TestTeeModel(t *testing.T) {
initChainFB(t)
ms2, e := NewSet(hmm0, hmm1)
fatalIf(t, e)
testScorer := func() scorer {
return scorer{op: []float64{math.Log(0.4), math.Log(0.2), math.Log(0.4)}}
}
h2 := narray.New(3, 3)
h3 := narray.New(4, 4)
h2.Set(1, 0, 1)
h2.Set(0.5, 1, 1)
h2.Set(0.5, 1, 2)
h3.Set(.9, 0, 1)
h3.Set(.1, 0, 3) // entry to exit transition.
h3.Set(0.5, 1, 1)
h3.Set(0.5, 1, 2)
h3.Set(0.5, 2, 2)
h3.Set(0.5, 2, 3)
h2 = narray.Log(nil, h2.Copy())
h3 = narray.Log(nil, h3.Copy())
hmm2, err := ms2.NewNet("model 2", h2,
[]model.Modeler{nil, testScorer(), nil})
fatalIf(t, err)
hmm3, errr := ms2.NewNet("model 3", h3,
[]model.Modeler{nil, testScorer(), testScorer(), nil})
fatalIf(t, errr)
hmms2, err := ms2.chainFromNets(xobs, hmm0, hmm2, hmm3, hmm0, hmm0, hmm0, hmm0, hmm2)
if err != nil {
t.Fatal(err)
}
hmms2.update()
nq := hmms2.nq
alpha2 := hmms2.alpha.At(nq-1, hmms2.ns[nq-1]-1, nobs-1)
beta2 := hmms2.beta.At(0, 0, 0)
t.Logf("alpha2:%f", alpha2)
t.Logf("beta2:%f", beta2)
// check log prob per obs calculated with alpha and beta
delta := math.Abs(alpha2-beta2) / float64(nobs)
if delta > smallNumber {
t.Fatalf("alphaLogProb:%f does not match betaLogProb:%f", alpha2, beta2)
}
_, err = ms2.chainFromNets(xobs, hmm3, hmm0, hmm2, hmm2)
if err == nil {
t.Fatal("expected error, got nil - first model has trans from entry to exit")
}
_, err = ms2.chainFromNets(xobs, hmm1, hmm0, hmm2, hmm3)
if err == nil {
t.Errorf("expected error, got nil - last model has trans from entry to exit")
}
}
