SFrame

SFrame is a data frame–like data structure used in the GraphLab Create framework for scalable tabular data analysis. It stores data as a collection of named columns, with each column assigned a specific type (such as integer, float, string, or boolean). SFrame is designed to handle datasets larger than available memory by using out-of-core processing and parallel computation across multiple cores or machines.

The SFrame API is accessed primarily through Python and supports common data manipulation tasks. Users can create SFrames from sources such as CSV files or from existing SArray or SFrame objects, select and drop columns, filter rows, add derived or new columns, sort data, perform group-by aggregations, and join with other SFrames. The data is stored columnwise to optimize performance for typical analytics and feature engineering pipelines.

SArray and SFrame are integrated components of the GraphLab Create ecosystem. SArray provides one-dimensional arrays that back the columns of an SFrame, enabling columnar operations and efficient feature construction. Together, they support end-to-end workflows from data cleaning and transformation to model training within the same framework.

Persistence and interchange are supported through saving and loading SFrames to disk. SFrames can also be exported to common formats such as CSV for interoperability, while internal representations may use a binary or serialized GraphLab Create format to maximize speed for large-scale processing.

Common use cases for SFrame include preprocessing large datasets for machine learning, exploratory data analysis on big data, and building scalable data-processing pipelines that feed features into models within the GraphLab Create ecosystem.