Free Examples From One Sentence

Word2vec needs no labelling: it slides a window along the text and every position hands it another training example.

Take one sentence of exactly 14 tokens, and run the window like this:

• every token in turn acts as the centre
• the context is the tokens lying within 3 positions on each side of the centre
• each (centre, one context token) pair counts as one training example

The window never wraps around and never runs past the ends of the sentence, so a centre near an end simply has fewer context tokens available on that side.

How many training examples does this one sentence produce?

Your answer is a whole number.