Section 18 of 24 · 5 lessons

Graphs and Relational Learning

A pump's reading can depend on the network around it. Graphs let us record those connections and use neighbouring measurements as context. We will begin with the representation, then follow information through walks, aggregation, learned layers, and predictions about missing links.

0 of 5 finished

Representing Relationships

Independent rows leave out who is connected to whom. Record a network carefully, then follow probability through its links.

  1. 01Recording Which Things Are ConnectedA table of pump readings leaves out the pipes between pumps. A graph records those relationships so calculations can use the connections.
    Build with Botie →Read the lesson
  2. 02Following Probability Through a NetworkA random walker chooses among outgoing links. Follow the probability mass to see how graph structure changes where it accumulates.
    Build with Botie →Read the lesson

Learning Across Connections

A node's neighbours can supply context. Gather that information, transform it with shared weights, and test what it tells us about missing links.

  1. 03Letting a Node Read Its Neighbours’ MeasurementsA pump’s neighbours can provide useful context. Aggregate their readings while retaining the distinction between a node and its surroundings.
    Build with Botie →Read the lesson
  2. 04Learning From a Node and Its Neighbours TogetherA graph convolution combines node features, self-connections, degree scaling, and learned weights. Work through these pieces before calling the result a layer.
    Build with Botie →Read the lesson
  3. 05Scoring Connections We Have Not ObservedNode representations can score possible links, but missing links are not guaranteed negatives. Design the split and candidate set before evaluating.
    Build with Botie →Read the lesson