Weka - Python wrapper for Weka classifiers

Overview

Provides a convenient wrapper for calling Weka classifiers from Python.

Installation

First install the Weka and LibSVM Java libraries. On Debian/Ubuntu this is simply:

sudo apt-get install weka libsvm-java

Then install the Python package with pip:

sudo pip install weka

Usage

Train and test a Weka classifier by instantiating the Classifier class, passing in the name of the classifier you want to use:

from weka.classifiers import Classifier
c = Classifier(name='weka.classifiers.lazy.IBk', ckargs={'-K':1})
c.train('training.arff')
predictions = c.predict('query.arff')

Alternatively, you can instantiate the classifier by calling its name directly:

from weka.classifiers import IBk
c = IBk(K=1)
c.train('training.arff')
predictions = c.predict('query.arff')

The instance contains Weka's serialized model, so the classifier can be easily pickled and unpickled like any normal Python instance:

c.save('myclassifier.pkl')
c = Classifier.load('myclassifier.pkl')
predictions = c.predict('query.arff')

Development

Tests require the Python development headers to be installed, which you can install on Ubuntu with:

sudo apt-get install python-dev python3-dev python3.4-dev

To run unittests across multiple Python versions, install:

sudo apt-get install python3.4-minimal python3.4-dev python3.5-minimal python3.5-dev

To setup a local Python virtual environment for testing, run:

./init_virtualenv.sh

To run all tests:

export TESTNAME=; tox

To run tests for a specific environment (e.g. Python 3.9):

export TESTNAME=; tox -e py39

To run a specific test:

export TESTNAME=.test_IBk; tox -e py39