===================================================================== ====== README ====== MILK 1.0 26 February 2004 Multi-Instance Learning Kit in Java Copyright (C) 2001-2004 University of Waikato web: http://www.cs.waikato.ac.nz/~ml/milk ===================================================================== NOTE: This software is experimental and not fit for general consumption. ===================================================================== Contents: --------- 1. Installation 2. Getting started - Classifiers - Data format - Experiment GUI - Data generators - Visualization tool - Javadoc 3. Credits 4. Submission of code and bug reports 5. Copyright ---------------------------------------------------------------------- 1. Installation: ---------------- MILK requires WEKA 3.4.1 to run. Most of it should work with other versions of WEKA as well. To run MILK classes, simply include the directory that contains the "milk" directory in your classpath. ---------------------------------------------------------------------- 2. Getting started: ------------------- =========== Classifiers =========== Assuming you have a dataset in ARFF format called "data.arff", you can run a cross-validation for (a version of) the diverse density algorithm on it using: java milk.classifiers.DD -t data.arff This prints out the generated classifier and ten-fold cross-validation estimates of its performance. If you don't pass any options to the classifier, MILK will list all the available options. Try: java milk.classifiers.DD The options are divided into "general" options that apply to most classification schemes in MILK, and scheme-specific options that only apply to the current scheme---in this case DD. This is just the same as in WEKA. =========== Data format =========== Datasets for MILK should be formatted according to the ARFF format. See the WEKA documentation for more details on this format. The first attribute must be a nominal attribute containing the bag index and every bag must have a different index. ==== GUIs ==== There is a version of the WEKA Experimenter that can be used to run experiments on multi-instance data. You can call it using java milk.gui.experiment.MIExperimenter Note that the GenericObjectEditor.props file included in the MILK directory must be in the directory from where you start the Experimenter or your home directory for this to work. You can even run distributed experiments! For more info, check the corresponding distribution for the WEKA Experimenter. =============== Data generators =============== There are some data generators, etc. provided in the milk.data package. ================== Visualization tool ================== A tool for visualizing multi-instance data is in milk.visualize. Try: java milk.visualize.MIExplorer ======= Javadoc ======= The Javadoc generated from the source code is in the doc directory. ---------------------------------------------------------------------- 3. Credits: ----------- Most of the code has been written (or adapted from WEKA code) by Xin Xu. The Experimenter has been adapated by Eibe Frank. Most of the classifiers in MILK are described in Xin Xu's MSc thesis. ---------------------------------------------------------------------- 4. Call for code and bug reports: --------------------------------- If you want to have your code included in MILK or you find some bugs, please send email to eibe@cs.waikato.ac.nz. ----------------------------------------------------------------------- 5. Copyright: ------------- MILK is distributed under the GNU public license. Please read the file COPYING. -----------------------------------------------------------------------