Part 32 · Module 4: Deep Learning Basics

Lab: Train the ticket network and explore embeddings

17 min read·13 Oct 2026

Module 4 · Lab · About 60 minutes · Needs: lessons 1 to 3 (the training loop, embeddings, TicketNet and the DataLoader); no model key needed

What you are building

The finished piece: the lessons' split feeds a properly trained network (blue, your system), a fair comparison with Module 3's classifier (green, the check), an embedding store (amber, the model's output) and explore(text), ending in two files that reload.

Module 4 adds two things to BrightCart: a small neural network that sorts tickets, and a tool for looking around the embedding space. This lab assembles the lessons' pieces, trains the network properly, and puts a number next to Module 3's scikit-learn classifier on the same 8 held-out tickets. Then it builds explore(text), a function Priya can hand any ticket to and get back the category the network thinks, the nearest tickets already answered, and the nearest help articles. It ends with two saved files, ticket_net.pt and brightcart_embeddings.npz, that reload in a fresh session and give the same answers. The split is lesson 1's, the network lesson 3's and the vectors lesson 2's; every number is printed.

What you are building

The rest of this course is yours to keep