开源软件名称：

开源软件地址：

开源软件介绍：

sk-dist: Distributed scikit-learn meta-estimators in PySpark

What is it?

sk-dist is a Python package for machine learning built on top ofscikit-learn and isdistributed under the Apache 2.0 softwarelicense. Thesk-dist module can be thought of as "distributed scikit-learn" asits core functionality is to extend the scikit-learn built-injoblib parallelization of meta-estimator training tospark. A popular use case is theparallelization of grid search as shown here:

Check out the blog postfor more information on the motivation and use cases of sk-dist.

Main Features

• Distributed Training - sk-dist parallelizes the training ofscikit-learn meta-estimators with PySpark. This allowsdistributed training of these estimators without any constraint onthe physical resources of any one machine. In all cases, sparkartifacts are automatically stripped from the fitted estimator. Theseestimators can then be pickled and un-pickled for prediction tasks,operating identically at predict time to their scikit-learncounterparts. Supported tasks are:
  • Grid Search: Hyperparameter optimizationtechniques,particularlyGridSearchCVandRandomizedSeachCV,are distributed such that each parameter set candidate is trainedin parallel.
  • Multiclass Strategies: Multiclass classificationstrategies,particularlyOneVsRestClassifierandOneVsOneClassifier,are distributed such that each binary probelm is trained inparallel.
  • Tree Ensembles: Decision treeensemblesfor classification and regression, particularlyRandomForestandExtraTrees,are distributed such that each tree is trained in parallel.
• Distributed Prediction - sk-dist provides a prediction modulewhich builds vectorizedUDFsforPySparkDataFramesusing fitted scikit-learn estimators. This distributes thepredict and predict_proba methods of scikit-learnestimators, enabling large scale prediction with scikit-learn.
• Feature Encoding - sk-dist provides a flexible featureencoding utility called Encoderizer which encodes mix-typedfeature spaces using either default behavior or user definedcustomizable settings. It is particularly aimed at text features, butit additionally handles numeric and dictionary type feature spaces.

Installation

Dependencies

sk-dist requires:

• Python (>= 3.5)
• scikit-learn (>=0.20.0,<0.23.2)
• pandas (>=0.17.0)
• numpy
• scipy
• joblib

Dependency Notes

• versions of numpy, scipy and joblib that are compatible with any supported version of scikit-learn should be sufficient for sk-dist
• sk-dist is not supported with Python 2

Spark Dependencies

Most sk-dist functionality requires a spark installation as well asPySpark. Some functionality can run without spark, so spark relateddependencies are not required. The connection between sk-dist and sparkrelies solely on a sparkContext as an argument to varioussk-dist classes upon instantiation.

A variety of spark configurations and setups will work. It is left up tothe user to configure their own spark setup. The testing suite runsspark 2.4 and spark 3.0, though any spark 2.0+ versionsare expected to work.

Additional spark related dependecies are pyarrow, which is used onlyfor skdist.predict functions. This uses vectorized pandas UDFs whichrequire pyarrow>=0.8.0, tested with pyarrow==0.16.0.Depending on the spark version, it may be necessary to setspark.conf.set("spark.sql.execution.arrow.enabled", "true") in thespark configuration.

User Installation

The easiest way to install sk-dist is with pip:

pip install --upgrade sk-dist

You can also download the source code:

git clone https://github.com/Ibotta/sk-dist.git

Testing

With pytest installed, you can run tests locally:

pytest sk-dist

Examples

The package contains numerousexampleson how to use sk-dist in practice. Examples of note are:

• Grid Search with XGBoost
• Spark ML Benchmark Comparison
• Encoderizer with 20 Newsgroups
• One-Vs-Rest vs One-Vs-One
• Large Scale Sklearn Prediction with PySpark UDFs

Gradient Boosting

sk-dist has been tested with a number of popular gradient boosting packages that conform to the scikit-learn API. Thisincludes xgboost and catboost. These will need to be installed in addition to sk-dist on all nodes of the sparkcluster via a node bootstrap script. Version compatibility is left up to the user.

Support for lightgbm is not guaranteed, as it requires additional installations on allnodes of the spark cluster. This may work given proper installation but has not beed tested with sk-dist.

Background

The project was started at IbottaInc. on the machine learningteam and open sourced in 2019.

It is currently maintained by the machine learning team at Ibotta. Specialthanks to those who contributed to sk-dist while it was initiallyin development at Ibotta:

• Evan Harris
• Nicole Woytarowicz
• Mike Lewis
• Bobby Crimi

Thanks to James Foley for logo artwork.