Extremely Random Trees

• What
• Building
• Usage
• Example: Growing a Model
• Example: Applying a Model
• Advanced Options

What

Extremely Random Trees is a machine-learning algorithm described in:

"Extremely randomized trees", DOI 10.1007/s10994-006-6226-1,
by Pierre Geurts, Damien Ernst, Louis Wehenkel, 2005

This is my own implementation of that algorithm as a friendly terminal program. It can be compiled and executed on any modern Unix/Linux/OSX computer.

Building

I've provided a Makefile which should do the job. After you've downloaded this repository, get into its directory in a terminal and do:

make

Usage

1. You provide training data.

• Training data is a list of sets of values.
• One set of values is one or more independent variables, and exactly one dependent variable.

2. You format the data into an ASCII (text) file following the comma-separated-value format (CSV).

• The first line of the file is an ordered list of variable names.
• Each following line represents one example, which is a complete set of values in the same order as the variable names.

3. You run etgrow using that file as a training file, and specifying where you'd like the output model to go.
4. If you have more data for which the dependent variable is unknown, then you run etpredict using the model from etgrow to make predictions about the values of the dependent variable.

Example: Growing a Model

./etgrow -t data/spambase-train.csv -m spambase-model

spambase-train.csv is an example input training file, provided in this repository.
spambase-model is the output model file that etgrow will create, and can be named whatever you want.

Example: Applying a Model

./etpredict -m spambase.model -t data/spambase-test.csv -p spambase.predictions

spambase.model is a model file that you created with etgrow at some point.
spambase-test.csv is an example input testing file, provided in this repository.
spambase.predictions is the output predictions file that etpredict will create, and can be named whatever you want. It contains an ordered list of dependent variable values.

Advanced Options

etgrow uses reasonable default values for the algorithm hyperparameters, but it is also capable of finding the best values for all of the hyperparameters if you have a few minutes to let it think.
To see a complete list of program options, just run etgrow with no command-line parameters.