Turn One-Hot Inputs Into Embeddings
Explain why embeddings replace sparse one-hot categories with dense learned vectors.
A recommendation model uses product_id with 80,000 possible values. Product A is a running shoe, Product B is a trail shoe, and Product C is a laptop stand. Embedding = learned dense vector for a category, updated during training so useful similarities can emerge. The common trap is saying embeddings understand categories. They encode learned similarity for the task, which may or may not match human meaning. One-hot baseline Product A = one position set to 1 in an 80,000-length vector The representation is sparse and treats every product as equally unrelated. Embedding lookup Product A maps to a short…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in