FROM DATAFRAMES TO GRAPH Data Science with pyTigerGraph
0 likes120 views
The document introduces PytigerGraph, a Python package that facilitates the creation, loading, and analysis of graph relationships, catering to data scientists familiar with Python. It discusses the advantages of using graph machine learning algorithms in various applications including social networks and fraud detection, while highlighting the trade-offs between Python and GSQL. Additionally, it emphasizes the importance of enabling graph data science and the future directions for leveraging graph technologies.
FROM DATAFRAMES TO GRAPH Data Science with pyTigerGraph
- 1. FROM DATAFRAMES TO GRAPH Data Science with pyTigerGraph Parker Erickson Graph+AI World 2020
- 2. ABOUT ME • Senior at University of Minnesota pursuing a B.S. and M.S. in Computer Science • Creator of pyTigerGraph • Part time software engineer at Optum • Patent-pending inventor for a fraud detection algorithm using graph ML • Connect with me: https://www.linkedin.com/in/parker-erickso n/ PARKER ERICKSON ©2020
- 3. OBJECTIVES • Learn about pyTigerGraph Python package • What it is used for • Learn why Data Scientists prefer to use Python • Intro to Graph ML Algorithms • Learn about when to use Python or GSQL • Tradeoffs between them • Learn how to get started with pyTigerGraph • Future Direction PARKER ERICKSON ©2020
- 4. WHY GRAPH? • Graphs not only store data, they store relationships between data in things like: • Social Networks • Fraud Rings • Recommendation Engines • Graph Machine Learning algorithms are new (~2016) • Graphs can enable Explainable AI https://tech.ebayinc.com/research/explainable-reasoning-over-knowledge-graphs-for-recommendation/ PARKER ERICKSON ©2020
- 5. ENABLING GRAPH DATA SCIENCE • Many organizations have data scientists that are Python experts • Tools should be what they are comfortable with • Lower need for simple GSQL queries • Loading data, simple analysis • Focus on algorithms and analysis, not learning new languages PARKER ERICKSON ©2020
- 6. PYTIGERGRAPH • pyTigerGraph enables data scientists to easily create, load, and analyze graph relationships • Opens the door to various graph machine learning algorithms, such as: • Node2Vec • Graph Convolutional Neural Networks • Graph Attention Networks PARKER ERICKSON ©2020
- 7. TRADITIONAL ML + GRAPH • Graph can accelerate traditional ML workflows • Lack of costly JOINs across tables accelerate development • Enrich knowledge graphs using outputs from ML algorithms • Sentiment Analysis • Entity Extraction PARKER ERICKSON ©2020
- 8. GRAPH MACHINE LEARNING • Node2Vec • Creates embeddings based on random walks • Based on Word2Vec • Graph Neural Networks • Uses “message passing” to generate representations of vertices in graph • Used for classification and regression tasks • Graph Convolutional Neural Networks (Kipf & Welling, 2017) • Graph Attention Networks (Veličković et al., 2018) • https://parkererickson.github.io/graph- https://app.wandb.ai/yashkotadia/gatedgcn-pattern/reports/Part-1-Introduction-to-Graph-Neural-Networks-with-GatedGCN--V mlldzoyMDg4MjA PARKER ERICKSON ©2020
- 10. CONTRIBUTE • Special thanks to: • Jon Herke – TigerGraph • Szilard Barany – TigerGraph • Yaniv Ben-Ami – Carleton College • Submit an Issue or Pull Request: • https://github.com/pyTigerGraph/pyTigerGra ph PARKER ERICKSON ©2020
- 11. QUESTIONS • Contact me at: • [email protected] • https://www.linkedin.com/in/parker-erickson/ • Notebooks are Here: • https://github.com/parkererickson/graphAIWorldDataframeToGraph • Connect with the Community on Discord: • https://discord.gg/XM7Cn9w PARKER ERICKSON ©2020