Implementations of Online Random Forest [1] in Scala and Python.
Online random forests are ensembles of online decision trees. Online decision trees are updated when new observations are presented. The observations need not be stored after they are used to update the trees / forest.
In the sbt directory in the terminal, execute
sbt package. This will generate orf_2.10-versionNumber.jar in
sbt/scala/target/scala-2.10/. To use it in a scala session, execute
scala -cp orf_2.10-versionNumber.jar. Import in scala using
import ORF.models._ .
Before testing, make sure to download the usps dataset. Go to
sbt/src/test/resources/usps and then execute ./download_data. After this,
the tests provided should executed properly by sbt ~test in the sbt
directory. The test currently takes about a minute.
See README.md in the sbt/ directory.
See README.md in the python/ directory.
[1] Amir Saffari, Christian Leistner, Jakob Santner, Martin Godec, and Horst Bischof, "On-line Random Forests," 3rd IEEE ICCV Workshop on On-line Computer Vision, 2009.