Phase 3 · Machine learning · Models
Multiclass classification
Softmax and cross-entropy for more than two classes.
Open this lesson Create a free account
Why it matters
Many questions have more than two answers: which product category a customer prefers, which team should handle a support ticket, which of five intents a chat message expresses. The tool is the same idea as logistic regression with softmax on top, and it's exactly the last layer of a language model, which is a classifier over roughly 100,000 tokens. The confusion matrix you'll learn here is how you find out what a classifier confuses, not just how often it's wrong.
What you'll be able to do
- Extend logistic regression to several classes with softmax
- Train a softmax classifier from scratch with one-hot labels
- Read a confusion matrix to see which classes get mixed up
- Compare against a majority baseline and per-class results
Before this lesson
Next in this phase: Decision trees