TigerGraph unveils new tool for machine learning modeling
The vendor's new capability is a pipeline for developing and deploying machine learning models with graph technology and works with popular third-party machine learning platforms.
TigerGraph on Wednesday unveiled a new machine learning capability designed to speed the development and improve the accuracy of data science models.
The graph database vendor, founded in 2012 and based in Redwood City, Calif., unveiled TigerGraph ML Workbench in preview during Graph + AI Summit, the spring edition of the biannual open conference hosted virtually by TigerGraph.
The tool, which is compatible with TigerGraph 3.2 and versions after that and can be deployed as either a fully managed cloud service or on premises, will be generally available in June 2022, according to the vendor.
TigerGraph's platform is based on graph technology, which enables data points within databases to simultaneously connect to multiple other data points rather than to just one other data point at a time, as within a traditional relational database. By connecting to multiple data points at once, users can more easily discover relationships between data, resulting in speed and accuracy, according to the vendor.
Neo4j is another specialist in graph database technology, while tech giants such as AWS, IBM, Microsoft and Oracle also offer graph database options.
Meanwhile, common use cases for graph technology include fraud detection and the development of social media platforms, both of which rely heavily on discovering relationships.
TigerGraph ML Workbench was built to make data science -- in particular, developing machine learning models -- faster, easier and more accurate in much the same way graph technology is speeding up the analytics process and enabling users to reach insight and action more quickly and accurately than they could with relational databases, according to Victor Lee, vice president of machine learning and AI at TigerGraph.
The tool enables users to develop machine learning models using graph neural networks -- data points simultaneously linked to multiple other data points rather than just one other data point -- and uses Jupyter Notebooks written in Python to enable the modeling of the entire machine learning workflow.
Jupyter Notebooks are open source web applications data scientists use to build and share their work.
"Machine Learning Workbench is designed to enable data scientists to easily use graph with their traditional machine learning techniques, while also getting the added benefit of what graph can do that other databases can't," Lee said.
Customers could previously use TigerGraph to build machine learning models, but they had to develop their own machine learning pipelines on top of their graph database.
But with Jupyter Notebooks, ML Workbench provides the pipeline where data is curated and models are trained, according to Lee.
"The Workbench is putting the whole thing together," he said. "Data scientists use Jupyter Notebooks, they use Python, and the Workbench presents Jupyter Notebooks written in Python to model the entire graph-based machine learning workflow."
The target audience for TigerGraph ML Workbench is data scientists, but by providing the machine learning workflow in one tool, the vendor is attempting to make developing machine learning models on top of TigerGraph simpler, and therefore open it up to a wider array of organizations.
Most large enterprises have data science teams that can dedicate time to building predictive models, but many smaller organizations don't have the same resources. ML Workbench saves time and adds accuracy, making predictive modeling possible for anyone with data science skills, Lee said.
While ML Workbench advances the machine learning capabilities of TigerGraph's platform, a key aspect of the new tool is its openness, according to Doug Henschen, an analyst at Constellation Research.
ML Workbench was built to work in concert with deep learning frameworks including DGL (deep graph library), PyTorch, PyTorch Geometric and TensorFlow, and it can be integrated with the machine learning tools from major cloud service providers AWS (SageMaker), Google (Vertex AI) and Microsoft (Azure ML).
Doug HenschenAnalyst, Constellation Research
"TigerGraph's emphasis with this release is openness," Henschen said. "It's open to popular deep learning frameworks and API compatible" with the tech giants' major machine learning platforms.
"Data scientists who know what they are doing and who have clear preferences appreciate openness," he said.
TigerGraph ML Workbench advances the capabilities TigerGraph and other graph database vendors previously provided, he added.
In April 2021, TigerGraph unveiled prebuilt machine learning and data science algorithms for specific applications including fraud detection and cybersecurity. AWS subsequently launched Amazon Neptune ML in July, while Neo4j introduced Neo4j Graph Data Science in April 2020 to add prebuilt machine learning and data science algorithms.
"Having ML capabilities tied to your graph database seems to have become mandatory," Henschen said.
ML Workbench, however, takes predictive modeling further by providing the Python Notebook environment, he continued.
"With this release, TigerGraph is supplying a Python Notebook environment said to be compatible with a long list of popular open source libraries," Henschen said. "The prebuilt algorithms TigerGraph announced in 2021 covered use cases, so this announcement is opening things up to a general-purpose ML capability that could be used in many ways."
The version of TigerGraph ML Workbench that will be generally available next month is just the initial version, according to Lee.
Looking ahead, the vendor plans to make the tool easier to use so potential users who don't know Python -- or don't want to use Python -- can simply select options from a menu to build predictive models without writing code.
AWS already offers no-code predictive modeling capabilities, but not specific to graph databases.
In addition, future iterations of ML Workbench will include integrations with cloud platforms beyond AWS SageMaker, Google Vertex AI and Microsoft Azure ML, according to Lee.
"This is the initial version, but we have a roadmap planned out for technical improvements -- better performance, support for more platforms -- and making it easier to use," he said.
Making it easier to use ties in to TigerGraph's goal of making graph technology available to not only trained data scientists but also a broad audience of business users, he added.
"We want more people to enjoy the benefits of graph," Lee said. "[ML Workbench] was motivated by what we could do to democratize graph analytics and machine learning even more to take the mystery out of graph and so more people can see its benefits."