.. figure:: https://github.com/Ibotta/sk-dist/blob/master/doc/images/skdist.png :alt: sk-dist
|License| |PyPI Package| |Downloads| |Python Versions|
sk-dist is a Python package for machine learning built on top of
scikit-learn <https://scikit-learn.org/stable/index.html> and is
distributed under the Apache 2.0 software license <https://github.com/Ibotta/sk-dist/blob/master/LICENSE>. The
sk-dist module can be thought of as "distributed scikit-learn" as
its core functionality is to extend the scikit-learn built-in
joblib parallelization of meta-estimator training to
spark <https://spark.apache.org/>__. A popular use case is the
parallelization of grid search as shown here:
.. figure:: https://github.com/Ibotta/sk-dist/blob/master/doc/images/grid_search.png :alt: sk-dist
Check out the blog post <https://medium.com/building-ibotta/train-sklearn-100x-faster-bec530fc1f45>__
for more information on the motivation and use cases of sk-dist.
Distributed Training - sk-dist parallelizes the training of
scikit-learn meta-estimators with PySpark. This allows
distributed training of these estimators without any constraint on
the physical resources of any one machine. In all cases, spark
artifacts are automatically stripped from the fitted estimator. These
estimators can then be pickled and un-pickled for prediction tasks,
operating identically at predict time to their scikit-learn
counterparts. Supported tasks are:
Hyperparameter optimization techniques <https://scikit-learn.org/stable/modules/grid_search.html>,
particularly
GridSearchCV <https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV>
and
RandomizedSeachCV <https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.RandomizedSearchCV.html#sklearn.model_selection.RandomizedSearchCV>__,
are distributed such that each parameter set candidate is trained
in parallel.Multiclass classification strategies <https://scikit-learn.org/stable/modules/multiclass.html>,
particularly
OneVsRestClassifier <https://scikit-learn.org/stable/modules/generated/sklearn.multiclass.OneVsRestClassifier.html#sklearn.multiclass.OneVsRestClassifier>
and
OneVsOneClassifier <https://scikit-learn.org/stable/modules/generated/sklearn.multiclass.OneVsOneClassifier.html#sklearn.multiclass.OneVsOneClassifier>__,
are distributed such that each binary probelm is trained in
parallel.Decision tree ensembles <https://scikit-learn.org/stable/modules/ensemble.html#forests-of-randomized-trees>
for classification and regression, particularly
RandomForest <https://scikit-learn.org/stable/modules/ensemble.html#random-forests>
and
ExtraTrees <https://scikit-learn.org/stable/modules/ensemble.html#extremely-randomized-trees>__,
are distributed such that each tree is trained in parallel.Distributed Prediction - sk-dist provides a prediction module
which builds vectorized UDFs <https://spark.apache.org/docs/latest/sql-pyspark-pandas-with-arrow.html#pandas-udfs-aka-vectorized-udfs>
for
PySpark <https://spark.apache.org/docs/latest/api/python/index.html>
DataFrames <https://spark.apache.org/docs/latest/api/python/pyspark.sql.html#pyspark.sql.DataFrame>__
using fitted scikit-learn estimators. This distributes the
predict and predict_proba methods of scikit-learn
estimators, enabling large scale prediction with scikit-learn.
Feature Encoding - sk-dist provides a flexible feature
encoding utility called Encoderizer which encodes mix-typed
feature spaces using either default behavior or user defined
customizable settings. It is particularly aimed at text features, but
it additionally handles numeric and dictionary type feature spaces.
Dependencies
``sk-dist`` requires:
- `Python <https://www.python.org/>`__ (>= 3.5)
- `scikit-learn <https://scikit-learn.org/stable/>`__ (>=0.20.0,<0.23.2)
- `pandas <https://pandas.pydata.org/>`__ (>=0.17.0)
- `numpy <https://www.numpy.org/>`__
- `scipy <https://www.scipy.org/>`__
- `joblib <https://joblib.readthedocs.io/en/latest/>`__
Dependency Notesnumpy, scipy and joblib that are compatible with any supported version of scikit-learn should be sufficient for sk-distsk-dist is not supported with Python 2Spark Dependencies
Most ``sk-dist`` functionality requires a spark installation as well as
PySpark. Some functionality can run without spark, so spark related
dependencies are not required. The connection between sk-dist and spark
relies solely on a ``sparkContext`` as an argument to various
``sk-dist`` classes upon instantiation.
A variety of spark configurations and setups will work. It is left up to
the user to configure their own spark setup. The testing suite runs
``spark 2.4`` and ``spark 3.0``, though any ``spark 2.0+`` versions
are expected to work.
Additional spark related dependecies are ``pyarrow``, which is used only
for ``skdist.predict`` functions. This uses vectorized pandas UDFs which
require ``pyarrow>=0.8.0``, tested with ``pyarrow==0.16.0``.
Depending on the spark version, it may be necessary to set
``spark.conf.set("spark.sql.execution.arrow.enabled", "true")`` in the
spark configuration.
User InstallationThe easiest way to install sk-dist is with pip:
::
pip install --upgrade sk-distYou can also download the source code:
::
git clone https://github.com/Ibotta/sk-dist.gitTesting
With ``pytest`` installed, you can run tests locally:
::
pytest sk-dist
Examples
--------
The package contains numerous
`examples <https://github.com/Ibotta/sk-dist/tree/master/examples>`__
on how to use ``sk-dist`` in practice. Examples of note are:
- `Grid Search with XGBoost <https://github.com/Ibotta/sk-dist/blob/master/examples/search/xgb.py>`__
- `Spark ML Benchmark Comparison <https://github.com/Ibotta/sk-dist/blob/master/examples/search/spark_ml.py>`__
- `Encoderizer with 20 Newsgroups <https://github.com/Ibotta/sk-dist/blob/master/examples/encoder/basic_usage.py>`__
- `One-Vs-Rest vs One-Vs-One <https://github.com/Ibotta/sk-dist/blob/master/examples/multiclass/basic_usage.py>`__
- `Large Scale Sklearn Prediction with PySpark UDFs <https://github.com/Ibotta/sk-dist/blob/master/examples/predict/basic_usage.py>`_
Gradient Boosting
-----------------
``sk-dist`` has been tested with a number of popular gradient boosting packages that conform to the ``scikit-learn`` API. This
includes ``xgboost`` and ``catboost``. These will need to be installed in addition to ``sk-dist`` on all nodes of the spark
cluster via a node bootstrap script. Version compatibility is left up to the user.
Support for ``lightgbm`` is not guaranteed, as it requires `additional installations <https://lightgbm.readthedocs.io/en/latest/Installation-Guide.html#linux>`__ on all
nodes of the spark cluster. This may work given proper installation but has not beed tested with ``sk-dist``.
Background
----------
The project was started at `Ibotta
Inc. <https://medium.com/building-ibotta>`__ on the machine learning
team and open sourced in 2019.
It is currently maintained by the machine learning team at Ibotta. Special
thanks to those who contributed to ``sk-dist`` while it was initially
in development at Ibotta:
- `Evan Harris <https://github.com/denver1117>`__
- `Nicole Woytarowicz <https://github.com/nicolele>`__
- `Mike Lewis <https://github.com/Mikelew88>`__
- `Bobby Crimi <https://github.com/rpcrimi>`__
Thanks to `James Foley <https://github.com/chadfoley36>`__ for logo artwork.
.. figure:: https://github.com/Ibotta/sk-dist/blob/master/doc/images/ibottaml.png
:alt: IbottaML
.. |License| image:: https://img.shields.io/badge/License-Apache%202.0-blue.svg
:target: https://opensource.org/licenses/Apache-2.0
.. |Build Status| image:: https://travis-ci.com/Ibotta/sk-dist.png?branch=master
:target: https://travis-ci.com/Ibotta/sk-dist
.. |PyPI Package| image:: https://badge.fury.io/py/sk-dist.svg
:target: https://pypi.org/project/sk-dist/
.. |Downloads| image:: https://pepy.tech/badge/sk-dist
:target: https://pepy.tech/project/sk-dist
.. |Python Versions| image:: https://img.shields.io/pypi/pyversions/sk-dist
:target: https://pypi.org/project/sk-dist/