pmmlTransformations (version 1.3.3)

pmmlTransformations-package: Data Transformations for PMML output

Description

NOTE: The functions in this package have been merged into the package pmml, starting with pmml 2.0.0. pmmlTransformations still works with pmml up to and including version 1.5.7, but will not receive any more updates. The examples throughout this package have (commented-out) calls to functions from pmml; if using pmmlTransformations, please use pmml 1.5.7 or older.

This package reads in raw data and allows the user to perform various transformations on the input data. The user can then use the derived data to construct models and output the model together with data transformations in the Predictive Model Markup Language (PMML) format through the use of the pmml package.

PMML is an XML-based language which provides a way for applications to define machine learning, statistical and data mining models and to share models between PMML compliant applications. More information about the PMML industry standard and the Data Mining Group can be found at <http://www.dmg.org>. The generated PMML can be imported into any PMML consuming application, such as Zementis Predictive Analytics products, which integrate with web services, relational database systems and deploy natively on Hadoop in conjunction with Hive, Spark or Storm, as well as allow predictive analytics to be executed for IBM z Systems mainframe applications and real-time, streaming analytics platforms.

Details

The general methodology to use this package is to first wrap the data with the WrapData function and then perform all desired transformations. The model, in PMML format, including the information on the transformations executed, can then be output by calling the pmml function of the pmml package. The pmml function in this case has to be given an additional parameter, transform, as shown in the example below.

This package can also be used as a transformation generator; output just the transformations information instead of the whole pmml model. To do so, one has to call the pmml function with the WrapData output but pass in a null value as the model name. An example can be seen in the documentation for the WrapData function.

This package does not support boolean dataTypes; only numeric and string data are supported.

References

PMML home page

PMML page describing various possible data transformations

A. Guazzelli, W. Lin, T. Jena (2012), PMML in Action: Unleashing the Power of Open Standards for Data Mining and Predictive Analytics. CreativeSpace (Second Edition) - Available on Amazon.com

T. Jena, A. Guazzelli, W. Lin, M. Zeller (2013). The R pmmlTransformations Package. In Proceedings of the 19th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

Examples

# NOT RUN {
# Load the standard iris dataset, already available in the base R package
   data(iris)

# First create the wrapper object
   irisBox <- WrapData(iris)

# Perform a simple z-transformation on the first variable of the dataset: 
# Sepal.Length. By default, the name of the transformed variable is 
# "derived_Sepal.Length". The information of the transformation is added 
# back to the wrapped data object.
   irisBox <- ZScoreXform(irisBox,"1")

# Build a simple lm model 
   fit <- lm(Sepal.Width ~ derived_Sepal.Length + Petal.Length, 
            data=irisBox$data)

# One may now output the model in PMML format using the command below.
# The PMML file will now include the data transformations as well as 
# the model. 
   # library(pmml)
   # fit_pmml <- pmml(fit, transform=irisBox)
# }