Vowpal Wabbit Python Wrapper

Vowpal Wabbit is a fast machine learning library for online learning, and this is the python wrapper for the project.

Installing this package builds Vowpal Wabbit locally for explicit use within python, it will not create the command-line version of the tool (or affect any previously existing command-line installations). To install the command-line version see the main project page: https://github.com/JohnLangford/vowpal_wabbit

The version of the PyPI vowpalwabbit package corresponds to the tagged version of the code in the github repo that will be used during building and installation. If you need to make local changes to the code and rebuild the python binding be sure to pip uninstall vowpalwabbit then rebuild using the local repo installation instructions below.

Vowpal Wabbit internally requires boost-python to be installed prior to installation. Install boost dependencies as below if they are not already installed.

$ sudo apt-get install libboost-program-options-dev zlib1g-dev libboost-python-dev

Installation

From PyPI:

$ pip install vowpalwabbit

From local repo (useful when making modifications):

$ cd python
$ pip install -e .

Usage

You can use the python wrapper directly like this:

>>> from vowpalwabbit import pyvw
>>> vw = pyvw.vw(quiet=True)
>>> ex = vw.example('1 | a b c')
>>> vw.learn(ex)
>>> vw.predict(ex)

Or you can use the included scikit-learn interface like this:

>>> import numpy as np
>>> from sklearn import datasets
>>> from sklearn.model_selection import train_test_split
>>> from vowpalwabbit.sklearn_vw import VWClassifier
>>>
>>> # generate some data
>>> X, y = datasets.make_hastie_10_2(n_samples=10000, random_state=1)
>>> X = X.astype(np.float32)
>>>
>>> # split train and test set
>>> X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=256)
>>>
>>> # build model
>>> model = VWClassifier()
>>> model.fit(X_train, y_train)
>>>
>>> # predict model
>>> y_pred = model.predict(X_test)
>>>
>>> # evaluate model
>>> model.score(X_train, y_train)
>>> model.score(X_test, y_test)

Troubleshooting

Some common causes of failure for installation are due to missing or mis-matched dependencies when Vowpal Wabbit builds. Make sure you have boost and boost-python installed on your system.

For Ubuntu/Debian/Mint

$ apt-get install libboost-program-options-dev zlib1g-dev libboost-python-dev

For Mac OSX

$ brew install libtool autoconf automake
$ brew install boost
$ brew install boost-python
# or for python3 (you may have to uninstall boost and reinstall to build python3 libs)
$ brew install boost-python3

Installing Vowpal Wabbit under an Anaconda environment (on OSX or Linux) can be done using the following steps:

git clone https://github.com/JohnLangford/vowpal_wabbit.git
# create conda environment if necessary
conda create -n vowpalwabbit
source activate vowpalwabbit
# install necessary boost dependencies
conda install -y -c anaconda boost
pip install -e vowpal_wabbit/python

Development

Contributions are welcome for improving the python wrapper to Vowpal Wabbit.

1. Check for open issues or create one to discuss a feature idea or bug.
2. Fork the repo on Github and make changes to the master branch (or a new branch off of master).
3. Write a test in the python/tests folder showing the bug was fixed or feature works (recommend using pytest).
4. Make sure package installs and tests pass under all supported environments (this calls tox automatically).
5. Send the pull request.

Tests can be run using setup.py:

$ python setup.py test

Directory Structure:

• python : this is where the c++ extension lives
• python/vowpalwabbit : this is then main directory for python wrapper code and utilities
• python/examples : example python code and jupyter notebooks to demonstrate functionality
• python/tests : contains all tests for python code

Note: neither examples nor tests directories are included in the distributed package, they are only for development purposes.