Appendix F. Using DataVec
DataVec is a library for handling machine learning data. DataVec handles the Extract, Transform, and Load (ETL) or vectorization component of a machine learning pipeline. The goal of DataVec is to simplify the preparation and loading of raw data into a format ready for use for machine learning. DataVec includes functionality for loading tabular (comma-separated values [CSV] files, etc.), image, and time-series datasets, both for single machine and distributed (Apache Spark) applications.
ND4J Vector Creation and DataVec
DataVec is meant to handle many of the feature and label creation chores mentioned previously in this book. Using DataVec is considered a best practice for DL4J workflows on a single machine and on Spark.
DataVec provides two main categories of functionality:
Functionality for loading data, from a variety of formats
Functionality for performing common data transformation operations (often called data wrangling or data munging)
These two categories of functionality are discussed separately in the sections that follow.
Loading Data for Machine Learning
Machine learning data comes in a wide variety of formats, with different requirements and libraries for loading each. Too often, machine learning practitioners end up writing one-off code to load their data; this can be both time consuming and error prone. DataVec attempts to alleviate these issues in two ways: first, by providing data loading functionality for common use ...