/spark-sklearn

Scikit-learn integration package for Spark

Primary LanguagePythonApache License 2.0Apache-2.0

Scikit-learn integration package for Apache Spark

This package contains some tools to integrate the Spark computing framework with the popular scikit-learn machine library. Among other tools:

  • train and evaluate multiple scikit-learn models in parallel. It is a distributed analog to the multicore implementation included by default in scikit-learn.
  • convert Spark's Dataframes seamlessly into numpy ndarrays or sparse matrices.
  • (experimental) distribute Scipy's sparse matrices as a dataset of sparse vectors.

It focuses on problems that have a small amount of data and that can be run in parallel.

  • for small datasets, it distributes the search for estimator parameters (GridSearchCV in scikit-learn), using Spark,

  • for datasets that do not fit in memory, we recommend using the distributed implementation in Spark MLlib.

    NOTE: This package distributes simple tasks like grid-search cross-validation. It does not distribute individual learning algorithms (unlike Spark MLlib).

Difference with the sparkit-learn project The sparkit-learn project aims at a comprehensive integration between Spark and scikit-learn. In particular, it adds some primitives to distribute numerical data using Spark, and it reimplements some of the most common algorithms found in scikit-learn.

License

This package is released under the Apache 2.0 license. See the LICENSE file.

Installation

This package is available on PYPI:

pip install spark-sklearn

This project is also available as as Spark package.

The developer version has the following requirements:

  • a recent release of scikit-learn. Releases 0.18.1, 0.19.0 have been tested, older versions may work too.
  • Spark >= 2.1.1. Spark may be downloaded from the Spark official website. In order to use this package, you need to use the pyspark interpreter or another Spark-compliant python interpreter. See the Spark guide for more details.
  • nose (testing dependency only)
  • Pandas, if using the Pandas integration or testing. Pandas==0.18 has been tested.

If you want to use a developer version, you just need to make sure the python/ subdirectory is in the PYTHONPATH when launching the pyspark interpreter:

PYTHONPATH=$PYTHONPATH:./python:$SPARK_HOME/bin/pyspark

Running tests You can directly run tests:

cd python && ./run-tests.sh

This requires the environment variable SPARK_HOME to point to your local copy of Spark.

Example

Here is a simple example that runs a grid search with Spark. See the Installation section on how to install the package.

from sklearn import svm, grid_search, datasets
from spark_sklearn import GridSearchCV
iris = datasets.load_iris()
parameters = {'kernel':('linear', 'rbf'), 'C':[1, 10]}
svr = svm.SVC()
clf = GridSearchCV(sc, svr, parameters)
clf.fit(iris.data, iris.target)

This classifier can be used as a drop-in replacement for any scikit-learn classifier, with the same API.

Documentation

API documentation is currently hosted on Github pages. To build the docs yourself, see the instructions in docs/README.md.

Changelog

  • 2015-12-10 First public release (0.1)
  • 2016-08-16 Minor release (0.2.0):
    1. the official Spark target is Spark 2.0
    2. support for keyed models
  • 2017-09-20 Minor release (0.2.2):
    1. The official Spark target is Spark >= 2.1
  • 2017-09-29 Minor release (0.2.3):
    1. Fixes spark-package build of spark-sklearn.