// Max returns the MAP state given a valuation.
func (g *Gaussian) Max(val VarSet) float64 {
v, ok := val[g.varid]
if ok {
z := math.Log(float64(C.gsl_ran_ugaussian_pdf(C.double((float64(v) - g.mean) / g.sd))))
return z
}
return math.Log(GaussMax)
}
// ArgMax returns both the arguments and the value of the MAP state given a certain valuation.
func (g *Gaussian) ArgMax(val VarSet) (VarSet, float64) {
retval := make(VarSet)
v, ok := val[g.varid]
if ok {
retval[g.varid] = v
z := math.Log(float64(C.gsl_ran_ugaussian_pdf(C.double((float64(v) - g.mean) / g.sd))))
return retval, z
}
retval[g.varid] = int(g.mean)
return retval, math.Log(GaussMax)
}
// Soft is a common base for all soft inference methods.
func (g *Gaussian) Soft(val VarSet, key string) float64 {
if _lv, ok := g.Stored(key); ok && g.stores {
return _lv
}
v, ok := val[g.varid]
var l float64
if ok {
l = float64(C.gsl_ran_ugaussian_pdf(C.double((float64(v) - g.mean) / g.sd)))
} else {
l = 1.0
}
g.Store(key, l)
return l
}
// LSoft is Soft in logspace.
func (g *Gaussian) LSoft(val VarSet, key string) float64 {
_lv := g.s[key]
if _lv >= 0 {
return _lv
}
v, ok := val[g.varid]
var l float64
if ok {
l = math.Log(float64(C.gsl_ran_ugaussian_pdf(C.double((float64(v) - g.mean) / g.sd))))
} else {
l = 0.0 // ln(1.0) = 0.0
}
g.Store(key, l)
return l
}
func UgaussianPdf(x float64) float64 {
return float64(C.gsl_ran_ugaussian_pdf(C.double(x)))
}