GraphLab

As the amounts of collected data and computing power grows (multicore, GPUs, clusters, clouds), modern datasets no longer fit into one computing node. Efficient distributed/parallel algorithms for handling large scale data are required. The GraphLab framework is a parallel programming abstraction targeted for sparse iterative graph algorithms. GraphLab provides a high level programming interface, allowing a rapid deployment of distributed machine learning algorithms.[3] The main design considerations behind the design of GraphLab are:

• Sparse data with local dependencies
• Iterative algorithms
• Potentially asynchronous execution

Main features of GraphLab are:

• A unified multicore and distributed API: write once run efficiently in both shared and distributed memory systems
• Tuned for performance: optimized C++ execution engine leverages extensive multi-threading and asynchronous IO
• Scalable: GraphLab intelligently places data and computation using sophisticated new algorithms
• HDFS Integration
• Powerful Machine Learning Toolkits

On top of GraphLab, several implemented libraries of algorithms:

• Topic modeling - contains applications like LDA which can be used to cluster documents and extract topical representations.[4]
• Graph analytics - contains application like pagerank and triangle counting which can be applied to general graphs to estimate community structure.[5]
• Clustering - contains standard data clustering tools such as Kmeans[6]
• Collaborative filtering - contains a collection of applications used to make predictions about users interests and factorize large matrices.[7]
• Graphical models - contains tools for making joint predictions about collections of related random variables.[8]
• Computer vision - contains a collection of tools for reasoning about images.[9]

A solution based on Graphlab collaborative filtering library won the 5th place in ACM Yahoo! KDD CUP challenge, track1, out of more than 1000 participants. LeBuShiShu team used a mixture of 12 different algorithms and deployed 10,000 CPU hours on BlackLight supercomputer.[10] Most of the utilized algorithms and techniques are now part of the GraphLab Collaborative FIltering Toolkit.

Turi (formerly called Dato and before that GraphLab Inc.) is a company that was founded by Prof. Carlos Guestrin from University of Washington in May 2013 to continue development support of the GraphLab open source project. Dato Inc. raised a $6.75M Series A from Madrona Venture Group and New Enterprise Associates (NEA). They raised a $18.5M Series B from Vulcan Capital and Opus Capital, with participation from Madrona and NEA.[11][12] On August 5, 2016, Turi was acquired by Apple Inc. for $200,000,000.[13][14]

• Turi homepage.
• SFrame: Extending GraphLab to tables