func TestDecisionTree(test *testing.T) {
x := db.NewNumericFeature(nil)
x.Add(0, 1, 2, 3, 4, 5, 6, 7)
y := db.NewNumericFeature(nil)
y.Add(1, 2, 1, 2, 1, 2, 1, 2)
z := db.NewNumericFeature(nil)
z.Add(1, 1, 1, 1, 2, 2, 2, 2)
t := db.NewNumericFeature(nil)
t.Add(3, 0, 3, 1, 7, 6, 5, -1)
dtFeatures := []db.OrderedFeature{}
for _, f := range []*db.NumericFeature{x, y, z} {
dtFeatures = append(dtFeatures, f)
}
dt := db.NewDecisionTreeRegressor()
dt.Fit(dtFeatures, t)
tEstimate := dt.Predict([]db.Feature{x, y, z})
if len(tEstimate) != t.Len() {
test.Errorf("dt.Predict() returned result of length: %d; expected %d", len(tEstimate), t.Len())
}
for i, te := range tEstimate {
if te != t.Value(i) {
test.Errorf("Row %d: predicted %v; actual %v", i, te, t.Value(i))
}
}
fmt.Printf("%v\n", dt.Importances())
}
func (handler *Handler) Header(header []string) {
handler.header = header
handler.columns = make([]*db.NumericFeature, len(header))
for i := range handler.columns {
handler.columns[i] = db.NewNumericFeature(nil)
}
}
func featureFactory(s string) db.DataFrameFeature {
var feature db.Feature
switch s[0] {
case 'C':
feature = db.NewCategoricalFeature(db.NewStringTable())
default:
feature = db.NewNumericFeature(nil)
}
return db.NewDataFrameFeature(s, feature)
}
func TestNumericFeature(t *testing.T) {
// Assign to Feature to ensure NumericFeature implements Feature
var f db.Feature = db.NewNumericFeature(nil)
// Assign it back so we can use its NumericFeature methods
var nf = f.(*db.NumericFeature)
for _, s := range []string{"3.0", "2.0", "non-numeric"} {
nf.AddFromString(s)
}
for _, v := range []float64{1.0, 4.0} {
nf.Add(v)
}
// Make sure integers work
nf.Add(5)
check := func(index int, expected float64) {
actual := nf.NumericValue(index)
altActual := nf.Value(index).(float64)
if actual != altActual && math.IsNaN(actual) != math.IsNaN(altActual) {
t.Errorf("NumericValue(%d) is %g; Value(%d) is %v", index, actual, index, altActual)
}
if actual != expected && math.IsNaN(actual) != math.IsNaN(expected) {
t.Errorf("Get(%d) got %g; expecting %g", index, actual, expected)
}
}
check(0, 3.0)
check(1, 2.0)
check(2, math.NaN())
check(3, 1.0)
check(4, 4.0)
check(5, 5.0)
ordercheck := func(index int, expected int) {
orderedIndex := nf.InOrder(index)
if orderedIndex != expected {
t.Errorf("Get(%d) got %d; expecting %d", index, orderedIndex, expected)
}
}
nf.Prepare()
ordercheck(0, 3)
ordercheck(1, 1)
ordercheck(2, 0)
ordercheck(3, 4)
ordercheck(4, 5)
ordercheck(5, 2)
if nf.Len() != 6 {
t.Errorf("Len() returned %d; expecting %d", nf.Len(), 5)
}
}
func TestRandomForest(test *testing.T) {
x := db.NewNumericFeature(nil)
x.Add(0, 1, 2, 3, 4, 5, 6, 7)
y := db.NewNumericFeature(nil)
y.Add(1, 2, 1, 2, 1, 2, 1, 2)
z := db.NewNumericFeature(nil)
z.Add(1, 1, 1, 1, 2, 2, 2, 2)
t := db.NewNumericFeature(nil)
// t.Add(3, 0, 3, 1, 7, 6, 5, -1)
t.Add(0, 0, 0, 0, 1, 1, 1, 1)
rfFeatures := []db.OrderedFeature{}
for _, f := range []*db.NumericFeature{x, y, z} {
rfFeatures = append(rfFeatures, f)
}
rf := db.NewRandomForestRegressor(10)
oob := rf.Fit(rfFeatures, t)
tEstimate := rf.Predict([]db.Feature{x, y, z})
if len(tEstimate) != t.Len() {
test.Errorf("rf.Predict() returned result of length: %d; expected %d", len(tEstimate), t.Len())
}
fmt.Printf("oob preds: %v\n", oob)
for i, te := range tEstimate {
fmt.Printf("predicted: %v; actual: %v\n", te, t.Value(i))
// if te != t.Value(i) {
// test.Errorf("Row %d: predicted %v; actual %v", i, te, t.Value///(i))
// }
}
fmt.Printf("feature importances (forest): %v\n", rf.Importances())
}
func TestDataFrame(t *testing.T) {
names := []string{"a", "b", "c", "d"}
data := [][]interface{}{
{"1.0", "2.0", "3.0", "4.0"},
{1.1, 2.2, 3.3, 4.4}}
expected := [][]float64{
{1.0, 2.0, 3.0, 4.0},
{1.1, 2.2, 3.3, 4.4}}
nf := db.NewDataFrame()
for i := 0; i < 4; i++ {
nf.AddFeature(db.NewDataFrameFeature(names[i], db.NewNumericFeature(nil)))
}
if nf.Length() != 0 {
t.Error("Length should be 0")
}
for _, v := range data {
nf.AddRow(v)
}
if nf.Length() != 2 {
t.Error("Length should be 2")
}
if nf.Width() != 4 {
t.Error("Columns() should return 4")
}
for i := 0; i < nf.Length(); i++ {
for j := 0; j < nf.Width(); j++ {
if nf.Get(i, j) != expected[i][j] {
t.Error(fmt.Sprintf("Cell (%d, %d): got %v; expected %v",
i, j, nf.Get(i, j), expected[i][j]))
}
}
}
for j := 0; j < nf.Width(); j++ {
if nf.ColumnName(j) != names[j] {
fmt.Println(nf.ColumnName(j))
t.Error(fmt.Sprintf(`Column name: got "%v"; expected "%v"`, nf.ColumnName(j), names[j]))
}
}
}
