func NewMSEAccumulator(minLeafSize int) *MSEAccumulator {
return &MSEAccumulator{
left: stats.NewVarianceAccumulator(),
right: stats.NewVarianceAccumulator(),
bestMetric: math.Inf(1),
previousFeatureValue: math.Inf(-1),
count: 0,
minLeafSize: minLeafSize,
}
}
func TestVarianceAccumulator(test *testing.T) {
a := stats.NewVarianceAccumulator()
x := []float64{1.0, 2.0, 0.0, 5.0}
mu := []float64{1.0, 1.5, 1.0, 2.0}
s := []float64{0.0, 0.5, 2.0, 14.0}
for i, x := range x {
sse, mean := a.Add(x)
if a.Count() != i+1 {
test.Errorf("Count() returned %d; expected %d", a.Count(), i+1)
}
if a.Value() != sse {
test.Errorf("Value() returned %v but Add() returned %v", a.Value(), sse)
}
if sse != s[i] {
test.Errorf("Expected sse of %v; got %v", s[i], sse)
}
if a.Variance() != s[i]/float64(i+1) {
test.Errorf("Expected sse of %v; got %v", s[i]/float64(i+1), a.Variance())
}
if mean != mu[i] {
test.Errorf("Expected mean of %v; got %v", s[i], mean)
}
}
for i := len(x) - 1; i >= 0; i-- {
sse, mean := a.Subtract(x[i])
if a.Count() != i {
test.Errorf("Count() returned %d; expected %d", a.Count(), i)
}
if i > 0 {
if a.Value() != sse {
test.Errorf("Value() returned %v but Add() returned %v", a.Value(), sse)
}
if sse != s[i-1] {
test.Errorf("Expected sse of %v; got %v", s[i-1], a.Value())
}
if mean != mu[i-1] {
test.Errorf("Expected mean of %v; got %v", mu[i-1], a.Value())
}
if a.Variance() != s[i-1]/float64(i) {
test.Errorf("Expected sse of %v; got %v", s[i]/float64(i+1), a.Variance())
}
} else {
if sse != 0.0 || a.Count() != 0 || a.Variance() != 0.0 {
test.Errorf("Error is %v; count is %v; MSE is %v but all should be zero", sse, a.Count(), a.Variance())
}
}
}
}
func dtInitialize(target Feature, bag Bag) ([]*DecisionTreeNode, []int) {
node := &DecisionTreeNode{feature: -1}
splittableNodeMembership := make([]int, bag.Len())
acc := stats.NewVarianceAccumulator()
for i := 0; i < bag.Len(); i++ {
splittableNodeMembership[i] = 0 // Root node
for j := 0; j < bag.Count(i); j++ {
acc.Add(target.NumericValue(i))
}
}
node.Metric = Metric{size: acc.Count(), prediction: acc.Mean()}
// nextSplittableNodes is next generation of splittableNodes.
// It is initialized here (and re-generated during each
// iteration) and contains pointers to nodes that are eligible
// for splitting during next generation.
initialSplittableNodes := []*DecisionTreeNode{node}
return initialSplittableNodes, splittableNodeMembership
}