Nearest neighbor vs cosine similarity

this is my lab Id: zpwxnkgmsulr
Question: idx = nearest_neighbor(tweet_embedding, document_vecs, 1)
idx2 = np.argmax(cosine_similarity(document_vecs, tweet_embedding))
idx, idx2

(array([3333]), 5202)

why is the result different?

nearest_neighbor() is calling cosine_similarity() one row of document_vecs at time. Versus calling cosine_similarity() on the entire document_vecs matrix at once. I didn’t work through the math by hand, but my intuition is that the linalg(norm()) is going to be different.