Overview of SAS pipefitter

What is SAS pipefitter?

The SAS pipefitter package provides a Python API for developing pipelines for data transformation and model fitting as stages of a repeatable workflow.

_images/pipeline.png

The package enables you to work with data in SAS to implement stages such as the following:

  • Impute missing values.
  • Select and engineer features.
  • Fit parameter estimates with decision trees, neural networks, and other machine learning techniques.
  • Use hyperparameter tuning to speed model selection.
  • Score data and assess models.

Another important feature of pipefitter is that it builds on the abilities of two other Python packages provided by SAS:

SWAT
A package that enables data transfer and analysis with SAS Cloud Analytic Services–the centerpiece of the SAS Viya platform–for in-memory analytics.
SASPy
A package that enables data transfer and analysis with SAS 9.4–business-proven software for analytics, data manipulation, and visualization.

The pipeline stages, such as estimating parameters with logistic regression are designed to run identically in SAS 9 though SASPy or in CAS through SWAT. The pipeline automatically adjusts to run the analysis where your data is by detecting the data set type. SASdata objects will run through SASPy, whereas CASTable objects run through SWAT.

See Getting started for examples.

Dependencies

SAS pipefitter is a pure Python package that supports Python 2.7 or 3.4+. However, you must install one or both of the following packages that are used to connect to SAS 9.4 or SAS Viya.

Package SAS Platfom Python Support
SASPy 2.x SAS 9.4 or higher Python 3.x
SWAT v1.1.0+ SAS Viya Python 2.7 or 3.4+

Read the package documentation for additional dependencies, supported platforms, and so on.

Tip

The most common combination that provides the greatest flexibility is to use Linux with 64-bit Python 3.4+.

Project status

Release 1.0.0 of SAS pipefitter represents an early stage of development. This release exploits only a small portion of what is possible with SAS 9.4 and SAS Viya.

For more information, see the following: