TIGERGRAPH MACHINE LEARNING WORKBENCH

Unlock Better and Faster Predictions with Machine Learning on Connected Data

The TigerGraph Machine Learning Workbench is a Python-based framework that accelerates development of graph-enhanced machine learning, which leverages the added insight from connected data and graph features for better predictions. Due to its accurate predictive power stemming from unique graph features and graph models, the Machine Learning Workbench enables organizations to unlock even better insights and greater business impact.

Development Framework for Graph Machine Learning

Enables data scientists to create graph neural network models and graph-enhanced models with production scale data.

Python-level Functions and Capabilities

Prepackaged Python libraries for graph data processing, graph feature engineering, subgraph sampling, data loading, and caching for out-of-DB training. No GSQL experience is required.

Compatible with Popular Machine Learning Frameworks

Work with the most popular machine learning frameworks in the market including PyTorch Geometric and DGL.

Plug-and-Play Ready Machine Learning

Flexible integration paths to works with your existing machine learning infrastructure on Amazon SageMaker, Google Vertex AI, and Microsoft Azure Machine Learning.

Versatile

The Machine Learning Workbench makes it easy for AI/ML practitioners to explore graph neural networks. It is fully interoperable with popular deep learning frameworks:

  • PyTorch Geometric
  • DGL

The Machine Learning Workbench is plug-and-play ready for Amazon SageMaker, Google Vertex AI, and Microsoft Azure ML. 

Benefits

Bigger Business Impact​

  • Deeper and better insights with GNNs to improve model accuracy
  • Proven across multiple industries

Ease of Development​

  • Python level development
  • Abstracted GSQL logic for graph data processing/manipulation
  • Faster development with Python library

Seamless Integration​

  • Compatible with PyG, and DGL
  • Works with your existing machine learning infrastructure on AWS, GCP, and Microsoft Azure

Production Data Scale​

  • Billion node/edges scale
  • Built-in sub-graph sampling reduces the powerful hardware requirement to train GNNs on large data sets

Performance

  • Professionally written, optimized GSQL queries for data processing
  • Leveraging our parallelized and distributed core engine

Download the TigerGraph Machine Learning Workbench Datasheet

Explore our Graph Data Science Options

Enterprise Edition

Free Developer Edition

The TigerGraph Machine Learning Workbench is designed to work with enterprise-level data. Users can easily train graph machine learning models even on a large graph without needing a powerful machine thanks to the following built-in capabilities:

    • Native integration with TigerGraph distributed storage and massively parallel processing engine for persisting your connected data and executing graph algorithms with parallel compute.
    • Graph-based partitioning to generate training/validation/testing for supervised graph machine learning models.
    • Graph-based batching for mini-batch training to improve performance and reduce hardware requirements.
    • Sub-graph sampling to support leading-edge graph neural network modeling techniques.
    • Graph-based data loader for link and node prediction applications for both homogeneous and heterogeneous graphs.

The TigerGraph Machine Learning Workbench is compatible with TigerGraph Enterprise Edition v3.2 onwards running on-premises or in the public cloud, right out of the box.

ML Workbench v1 (Dev Edition)

Compatibility

  • Graph Database: TigerGraph DB compatibility v3.2+
  • Cloud ML Platforms: Amazon SageMaker, Azure ML, GCP Vertex
  • PyG, DGL ML Framework compatibility

Installation Onboarding:

  • Docker image
  • MacOS and Linux installers
  • pip install & conda install

Built-in Python-level Capabilities:

  • Graph data partitioning
  • Graph data loading & export (http)
  • Subgraph sampling
  • Data batching
  • Graph feature generation
  • GNN: Node prediction support
  • GNN: Heterogeneous graph support
  • GNN: Link prediction support
  • GNN Inference with real-time data

Limitations:

  • Data export from graph DB via HTTP only
  • Data export size
  • No support for distributed parallel training
  • Community supportGNN ML libraries: PyTorch Geometric, D

ML Workbench v1 (Enterprise Edition)

Compatibility

  • Graph Database: TigerGraph DB compatibility v3.2+
  • Cloud ML Platforms: Amazon SageMaker, Azure ML, GCP Vertex
  • PyG, DGL ML Framework compatibility

Installation Onboarding:

  • Docker image
  • MacOS and Linux installers
  • pip install & conda install

Built-in Python-level Capabilities:

  • Graph data partitioning
  • Graph data loading & export (http)
  • Subgraph sampling
  • Data batching
  • Graph feature generation
  • GNN: Homogeneous graph support
  • GNN: Node prediction support
  • GNN: Heterogeneous graph support
  • GNN: Link prediction support
  • GNN Inference with real-time data

Enterprise Level Features:

  • Data export from graph DBvia both HTTP and Kafka → More reliable and efficient
  • No size limit on data export → Ready for enterprise-level scale
  • Ability to support distributed parallel training → Shorten learning best practices from the experts
  • TigerGraph standard support SLA with 12(PST)x5 → Better support

Unlock better and faster predictions on your connected data using the tools you already know.