siamese-lstm-for-sentence-similarity
siamese-lstm-for-sentence-similarity copied to clipboard
sentence embedding
hi, as the author say, the sentence embedding should be the output of the final step, it's conflict with the your code——reduce_sum
what's more, why you choose tf.clip_by_value to map the score from (0, -∞),and [1, 5] -> (0, 1)? is there any reason?