From a7417a5b7c187f903324a8e65218a21be9428787 Mon Sep 17 00:00:00 2001
From: caoying03 <caoying03@baidu.com>
Date: Sun, 8 Jan 2017 16:28:02 +0800
Subject: [PATCH] delete train.list and test.list

---
 label_semantic_roles/.gitignore      | 0
 label_semantic_roles/README.md       | 2 +-
 label_semantic_roles/data/test.list  | 1 -
 label_semantic_roles/data/train.list | 1 -
 4 files changed, 1 insertion(+), 3 deletions(-)
 create mode 100644 label_semantic_roles/.gitignore
 delete mode 100644 label_semantic_roles/data/test.list
 delete mode 100644 label_semantic_roles/data/train.list

diff --git a/label_semantic_roles/.gitignore b/label_semantic_roles/.gitignore
new file mode 100644
index 0000000..e69de29
diff --git a/label_semantic_roles/README.md b/label_semantic_roles/README.md
index 6c0735f..0b9deae 100644
--- a/label_semantic_roles/README.md
+++ b/label_semantic_roles/README.md
@@ -70,7 +70,7 @@ RNN 等价于一个展开地前向网络，通常人们会认为 RNN 在时间
 在绝大多数自然语言处理任务中，我们几乎总是能拿到整个句子。这种情况下，$t$ 时刻如果能够像获取历史信息一样，得到未来的信息，对序列学习任务会有很大的帮助。为了克服这一缺陷，我们可以设计一种双向循环网络单元，它的思想简单且直接：对上一节的栈式循环神经网络进行一个小小的修改，堆叠多个 LSTM 单元，让每一层 LSTM 单元分别以：正向、反向、正向 …… 的顺序学习上一层的输出序列。于是，从第 2 层开始， $t$ 时刻我们的 LSTM 单元便总是可以看到历史和未来的信息。图3 是基于 LSTM 的双向循环神经网络结构示意图。
 
 <p align="center">    
-<img src="./image/bidirectional_stack_lstm.png" width = "60%" height = "60%" align=center><br>
+<img src="./image/bidirectional_stacked_lstm.png" width = "60%" height = "60%" align=center><br>
 图3. 基于 LSTM 的双向循环神经网络结构示意图
 </p>
 
diff --git a/label_semantic_roles/data/test.list b/label_semantic_roles/data/test.list
deleted file mode 100644
index ec370e8..0000000
--- a/label_semantic_roles/data/test.list
+++ /dev/null
@@ -1 +0,0 @@
-./data/feature
diff --git a/label_semantic_roles/data/train.list b/label_semantic_roles/data/train.list
deleted file mode 100644
index ec370e8..0000000
--- a/label_semantic_roles/data/train.list
+++ /dev/null
@@ -1 +0,0 @@
-./data/feature
-- 
GitLab