This lesson implements a compact dialog satisfaction classifier for task-oriented support
dialogs. Each synthetic example is a short dialog summary paired with one of three satisfaction
levels: dissatisfied, neutral, or satisfied.
The lesson covers:
- deterministic synthetic dialog data with three satisfaction labels
- pooled-embedding text classification
- config, vocab, metrics, and checkpoint artifacts
Run locally with:
python -m tracks.nlp.lesson_38_toy_dialog_satisfaction_prediction.train --device cpu --epochs 1