sassoftware / python-pipefitter

The SAS pipefitter package provides a Python API for developing pipelines for data transformation and model fitting as stages of a repeatable machine learning workflow in either SAS v9 or SAS Viya.

SAS Pipefitter

The SAS Pipefitter project provides a Python API for developing machine learning pipelines. The pipelines are built from stages that perform variable transformation, parameter estimation, and hyperparameter tuning.

A key feature of pipefitter is that the pipelines can run with a SAS 9.4 platform as well as with SAS Cloud Analytic Services (CAS) in the SAS Viya platform.

This package builds on the SASPy and SAS SWAT packages. Those packages enable Python to work with SAS 9.4 and SAS Viya, respectively.

Installation

SAS pipefitter can be installed as follows with `pip`:

pip install pipefitter

Alternatively, you can install a specific release as follows:

pip install https://github.com/sassoftware/python-pipefitter/releases/download/vX.X.X/pipefitter-X.X.X.tar.gz

Dependencies

The package is a pure Python package and works with Python 2.7 or 3.4+.

However, the packages that enable Python to work with the different SAS platforms introduce dependencies of their own.

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+

Resources

Pipefitter documentation

SASPy

SAS SWAT for Python

Python

python-pipefitter's Issues

Getting started examples fails with SAS backend.

The doc example is here.
You get an invalid syntax error in the SAS code.

Hyperparameter tuning fails because of .ix

hyperparameter tuning fails in the provided example

hpt = HyperParameterTuning(
          estimator=DecisionTree(target='label', 
              inputs=['a'+str(i) for i in range(50)]),
          param_grid=param_grid, 
          cv=3)

hpt.gridsearch(casdata)

with the following error

'SASDataFrame' object has no attribute 'ix' error

the reason seems to be that .ix is deprecated

score method cannot be used in another CAS session

Issue

The score method fails when used in a CAS session that is different from the CAS session used for fit

Possible explanation:

The model table has a session scope.

Desired behavior

Be able to fit the model and score it in two different sessions.

The score method does not output predictions

Issue

The score method of models and pipeline does not output the actual predicted values

Actual behavior

score method returns a Series of metric results:

Suggestions:

Option 1:

Change the score method to output a table of the results.
For CAS backend, it can be the name of the casout table as an additional value in the results table

Option 2:

Keep the behavior of score method as is, and add a predict method (similar to sklearn syntax) that outputs a table of the results

AttributeError when using gridsearch hyperparameter tuning.

I'm following the example from the documentation here: link

Get the following error message when trying to run the gridsearch:
AttributeError: 'SASDataFrame' object has no attribute 'ix'

The "train_ct" table is in fact a CASTable and has the attribute 'ix'. As there are no other arguments, I don't know wether this is user error or a bug.