ARFF Files

ARFF files (Attribute-Relation File Format) are the most common format for data used in Weka. Each ARFF file must have a header describing what each data instance should be like. The attributes that can be used are as follows:

• Numeric

Real or integer numbers.

• Nominal

Nominal attributes must provide a set of possible values. For example:

@ATTRIBUTE class        {Iris-setosa,Iris-versicolor,Iris-virginica}

• String

Allows for arbitrary string values. Usually processed later using the StringToWordVector filter.

• Date

Allows for dates to be specified. As with Java’s SimpleDateFormat, this date can also be formatted; it will default to ISO-8601 format.

An example header can be seen as follows:

@RELATION iris

@ATTRIBUTE sepallength  NUMERIC
@ATTRIBUTE sepalwidth   NUMERIC
@ATTRIBUTE petallength  NUMERIC
@ATTRIBUTE petalwidth   NUMERIC
@ATTRIBUTE class        {Iris-setosa,Iris-versicolor,Iris-virginica}

Following the header each instance must be listed with the correct number of instances; if an attributes value for an instance is not known a ? can be used instead. The following shows an example of the set of instances in an ARFF file:

@DATA
5.1,3.5,1.4,0.2,Iris-setosa
4.9,3.0,1.4,0.2,Iris-setosa
4.7,3.2,1.3,0.2,Iris-setosa
4.6,3.1,1.5,0.2,Iris-setosa
5.0,3.6,1.4,0.2,Iris-setosa

Loading ARFF Files

Depending on the version of Weka being used different methods for loading ARFF files should be utilised.

Weka <3.5.5

The following sample code shows how to load an ARFF file:

import weka.core.Instances;
import java.io.BufferedReader;
import java.io.FileReader;
...
BufferedReader reader = new BufferedReader(new FileReader("data.arff"));
Instances data = new Instances(reader);
reader.close();
data.setClassIndex(data.numAttributes() - 1);

The class index shows what attribute should be used for classification. In most ARFF files this is the last attribute which is why it is set to data.numAttributes() - 1. If you are using a Weka function, such as buildClassifier, you must set the class index.

Weka >=3.5.5

In the latest version of Weka it is very easy to load an ARFF file. This method can also load CSV files and any other files Weka can understand.

import weka.core.converters.ConverterUtils.DataSource;
...
DataSource source = new DataSource("data.arff");
Instances data = source.getDataSet();
if (data.classIndex() == -1) {
  data.setClassIndex(data.numAttributes() - 1);
}

Loading from Database

Many databases can be used in Weka. Firstly, the DatabaseUtils.props file must be edited to match your database; specifically you must provide your database’s name, location, port and correct driver.

jdbcDriver=org.gjt.mm.mysql.Driver
jdbcURL=jdbc:mysql://localhost:3306/my_database

Then the database can be loaded by using some simple code.

import weka.core.Instances;
import weka.experiment.InstanceQuery;
...
InstanceQuery query = new InstanceQuery();
query.setUsername("user");
query.setPassword("pass");
query.setQuery("select * from mytable");
Instances data = query.retrieveInstances();

Some notes about loading from a database:

• Make sure the correct JDBC driver is in your classpath.

• If you are using Microsoft Access then the JDBC-ODBC-driver which comes with the JDK can be used.

• The InstanceQuery method converts VARCHAR to nominal attributes and TEXT to string attributes. A filter, such as NominalToString or StringToNormal, can convert the attributes back to their correct type.