Add support for trained forest/model serialisation and deserialisation. Options:
serialise tree node objects one by one, so the deserialised representation is a set of tree node objects (i.e take a tree and reconstruct the tree node by node)
walk the tree and serialise it by storing the split value followed by offset indexes to the left and right child branches (byte or value indexes within the serialised representation). The deserialised version of the tree is then an array of split points and indexes, and calling classify would only involve jumping through the array, rather than constructing tree node objects in memory. This would require a specialised implementation of Tree but would be more memory efficient.
Add support for trained forest/model serialisation and deserialisation. Options: