Join my FREE course Basics of Graph Neural Networks ([ Ссылка ])!
Code for this project: [ Ссылка ]
Discord Server: [ Ссылка ]
Chapters
- 00:00 - Running the notebook
- 00:56 - Overview of content
- 01:53 - Relational databases as graphs
- 04:00 - Tabular datasets as graphs
- 08:33 - Q: Better to have more nodes or more relationships?
- 09:15 - Q: Can we use multiple edge types?
- 09:30 - Q: How do you build similarity-based connections?
- 11:24 - Q: How to decide between representing a column as a node or node property?
- 13:00 - Configuring the notebook environment
- 14:29 - Looking at example fraud data to design the graph
- 22:33 - Creating the nodes
- 25:00 - Creating the edges
- 26:50 - Instantiating the graph
- 27:32 - Attaching labels to the nodes
- 30:00 - Visualizing the graph
- 35:07 - Constructing node features
- 36:22 - Baseline (non-graph) models
- 41:40 - Label propagation
- 44:19 - Graph neural network
- 46:51 - Handling featureless nodes
- 51:21 - Q: How to figure out if test nodes are connected to train nodes?
- 53:33 - Q: How to split into train/test
- 54:21 - An alternative graph structure and time-directed edges
- 57:07 - Q: On the interaction of time and graph structure
- 58:45 - Q: Should I remove strong features to troubleshoot the model?
