Prediction by trained model

// buildTestInstance constructs a new test instance using the structure of the training data
func buildTestInstance(template base.FixedDataGrid, features []float64, dummyLabel string) (base.FixedDataGrid, error) {
	original, ok := template.(*base.DenseInstances)
	if !ok {
		return nil, errors.New("template is not DenseInstances")
	}

	allAttrs := original.AllAttributes()
	if len(features)+1 != len(allAttrs) {
		return nil, errors.New("feature count mismatch")
	}

	inst := base.NewDenseInstances()
	for _, attr := range allAttrs {
		inst.AddAttribute(attr)
	}
	inst.AddClassAttribute(original.AllClassAttributes()[0])
	inst.Extend(1)

	for i, attr := range allAttrs {
		spec, _ := inst.GetAttribute(attr)
		switch a := attr.(type) {
		case *base.FloatAttribute:
			inst.Set(spec, 0, base.PackFloatToBytes(features[i]))
		case *base.CategoricalAttribute:
			inst.Set(spec, 0, a.GetSysValFromString(dummyLabel))
		default:
			return nil, errors.New("unsupported attribute type")
		}
	}
	return inst, nil
}

// predict uses the trained classifier to predict the label of input features
func predict(classifier base.Classifier, template base.FixedDataGrid, features []float64) (string, error) {
	inst, err := buildTestInstance(template, features, "dummy")
	if err != nil {
		return "", err
	}
	pred, err := classifier.Predict(inst)
	if err != nil {
		return "", err
	}
	return pred.RowString(0), nil
}

...
dataPath := "datasets/data.csv"
instances, err := base.ParseCSVToInstances(dataPath, true)
rf := ensemble.NewRandomForest(10, 3)
testFeatures := []float64{0.7, 2.1, 3.5}
label, err := predict(rf, instances, testFeatures)
...