Hands-On Notebooks

Notebook 0 - Intro to TigerGraph Cloud and Loading Data

In this section, we will cover what TigerGraph is and its massively parallel processing architecture. The unique architecture allows TigerGraph to run highly-performant, distributed, and scalable graph data science algorithms. We will then load a dataset of transactions to TigerGraph Cloud and familiarize ourselves with the TigerGraph ML Workbench Cloud.

Notebook 1 - Graph Data Exploration

Algorithms such as Louvain community detection [1] have been very good at helping discover fraudulent transactions within financial interaction graphs. Maximal independent set has been used for non-conflicting routing problems. Cosine similarity of graph neighborhoods have been used in recommendation and classification tasks. In this section, we will cover some large classes of graph data science algorithms, such as community detection, centrality, and similarity that can be executed efficiently within the TigerGraph database. Using these algorithms, we will begin to analyze and perform exploratory data analysis on the NFT transaction dataset.

Notebook 2 - Graph + Machine Learning

Combining traditional features as well as ones derived from the graph can be a powerful technique for improving the accuracy of machine learning algorithms without moving to graph neural networks. In this section, we will cover how to utilize the TigerGraph Graph Data Science Library and pyTigerGraph to enrich existing traditional machine learning models with graph data derived from algorithms like PageRank [7]. We will use the graph algorithms used in the section above to develop features for traditional machine learning algorithms such as XGBoost [2] to predict the selling price of NFTs in the network.

Notebook 3 - Graph Neural Networks

Graph Neural Network models have been exploding in popularity in recent years, yet there have not been great ways to store and query the data into subgraphs for training models such as GraphSAGE [5]. TigerGraph simplifies this process through our Machine Learning Workbench. Researchers and data scientists can now test new architectures on arbitrarily large datasets, using tools that they are already familiar with such as PyTorch Geometric and DGL. In this section, we will train and evaluate a Graph Neural Network with the NFT data stored in TigerGraph, incorporating both the visual feature data that describes the NFTs as well as the network information into our predictions.