Thanks for the video! l2_normalization maps the embeddings to a unit sphere, this is done so only the direction of the vectors matters, not the magnitude.
@connor-shorten
3 жыл бұрын
Ah, I see. Thank you so much for the clarification!
@cameron4814
3 жыл бұрын
@7:30 m,y guess about why they avoid using np.where, because it has to scan the entire table for that condition rather than using an index like a dictionary, (its slower.)
@AI_For_Scientists
2 жыл бұрын
Great video, thanks for sharing! One question, does implementing this on top of a network improve the classification acc of that model?
Пікірлер: 9