RandomFerns

This package provides an implementation of the very fast Random Ferns classifier / regressor as described in M. Özuysal et al., "Fast Keypoint Recognition in 10 Lines of Code", CVPR 2007.

Usage

For classification:

using RandomFerns
rf = RandFerns(data, labels)            # data ... D x N, labels ... 1 x N
prediction = predict(rf, testdata)      # testdata ... D x M, prediction ... 1 x M

Training data:

Learnt classification boundaries:

For this example, training takes about 25ms, and prediction 36ms.

For regression:

using RandomFerns
rf = RandFerns(data, targets, regression = true)
                                        # data ... D x N, targets ... T x N
prediction = predict(rf, testdata)      # testdata ... D x M, prediction ... T x M

Notice the typical over-smoothing behavior, due to the simplicity of the model:

Due to the fact that the cells are stored in a sparse fashion, it is possible to use a very high number of splits (240), which e.g. in this case yields a much better result:

Parameters

RandFerns(data, targets; regression = false, nsplits = 10, nferns = 32)

RandFerns has the following parameters:

• regression - set to true for regression
• nsplits - how many splits in feature space, yielding 2^nsplits cells
• nfers- how many ferns to train