From 989e19caf938efb9e7a5be4e24f1fb378a31ed68 Mon Sep 17 00:00:00 2001 From: Yibing Liu Date: Wed, 11 Oct 2017 09:47:12 +0000 Subject: [PATCH] fix typos in margin_rank_loss_op --- paddle/operators/margin_rank_loss_op.cc | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/paddle/operators/margin_rank_loss_op.cc b/paddle/operators/margin_rank_loss_op.cc index 16c9b20a26..5be61dfec3 100644 --- a/paddle/operators/margin_rank_loss_op.cc +++ b/paddle/operators/margin_rank_loss_op.cc @@ -75,13 +75,13 @@ turns out loss(X1, X2, Label) = max(0, -Label * (X1 - X2) + margin). The attribute `margin` involved here helps make the predictions more robust. -Denote the item ranked higher as the positive sample, otherwise negative -sample. If the score of the two samples statisfies +Denote the item ranked higher as the positive sample, otherwise the negative +sample. If the score of the two samples satisfies positive sample - negative sample < margin, -the pair of samples will contribute to the loss, which will backpropogate and -train the ranking model to enlarge the difference of the two score. +the pair of samples will contribute to the final loss, which will backpropogate +and train the ranking model to enlarge the difference of the two score. For batch input with size `batch_size`, `X1`, `X2` and `Label` all have the same shape [batch_size x 1]. -- GitLab