promptdojo_

Embeddings as features, not magic — step 7 of 7

Checkpoint

One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.

Evaluate the zero-training classifier like a real model. Write classify(vec) (nearest anchor by cosine), run it over the five labeled docs, print one line per doc — f"{text:<22} pred={pred:<8} {mark}" where mark is "ok" or "MISS" — and finish with f"accuracy: {...:.0%}". Embeddings as features means the whole chapter-39 evaluation playbook applies to them unchanged.

full-screen editor opens — close anytime to keep reading.