Extended Isolation Forest

This is a simple Go implementation for the Extended Isolation Forest described in this paper. It is used for detecting anomalies and outliers in multidimensional data. Extended Isolation Forest in turn is an improvement on the Isolation Forest described in this paper.

This library currently only supports float64 type (of any dimension).

The authors of the Extended Isolation Forest have created a Python version of the algorithm.

Usage

go get -u github.com/oskanberg/eif-go

Please refer to the godoc reference.

In this basic example, a forest is built from a set of random normally distributed 3D points:

data := make([][]float64, 100)
for i := range data {
    data[i] = []float64{rand.NormFloat64(), rand.NormFloat64(), rand.NormFloat64()}
}
f := eif.NewForest(data, eif.WithMaxTreeDepth(12), eif.WithTrees(100))
fmt.Println(f.Score([]float64{0.0, 0.0, 0.0}))       // 0.37800519781971176
fmt.Println(f.Score([]float64{1.0, 1.0, 1.0}))       // 0.4581319136519572
fmt.Println(f.Score([]float64{100.0, 100.0, 100.0})) // 0.6520795100807051

Note that eif.WithMaxTreeDepth(12), eif.WithTrees(100) are optional parameters.