From e9d2fba8f1c93ece99b6e2ad45c5237404af3f6d Mon Sep 17 00:00:00 2001 From: "A. Unique TensorFlower" Date: Mon, 29 May 2017 08:42:33 -0700 Subject: [PATCH] Fix comment describing ignore_longer_outputs_than_inputs. PiperOrigin-RevId: 157400110 --- tensorflow/core/ops/ctc_ops.cc | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tensorflow/core/ops/ctc_ops.cc b/tensorflow/core/ops/ctc_ops.cc index 3d8c5339356..1a69106d80b 100644 --- a/tensorflow/core/ops/ctc_ops.cc +++ b/tensorflow/core/ops/ctc_ops.cc @@ -77,8 +77,8 @@ ctc_merge_repeated: Scalar. If set to false, *during* CTC calculation repeated non-blank labels will not be merged and are interpreted as individual labels. This is a simplified version of CTC. ignore_longer_outputs_than_inputs: Scalar. If set to true, during CTC - calculation items have longer input sequences than output sequences - are ignored by returning zero-gradient for those items. + calculation, items that have longer output sequences than input sequences + are skipped: they don't contribute to the loss term and have zero-gradient. loss: A vector (batch) containing log-probabilities. gradient: The gradient of `loss`. 3-D, shape: `(max_time x batch_size x num_classes)`.