From 1d163ad15f7bd37799c7015024cbebb110680b95 Mon Sep 17 00:00:00 2001
From: Xinghai Sun <sunxinghai1216@gmail.com>
Date: Thu, 31 Aug 2017 12:22:27 +0800
Subject: [PATCH] Fixed a serious mistake of bidirectional simple rnn for DS2.

---
 cloud/pcloud_submit.sh |  4 ++--
 layer.py               | 16 ++++++++++------
 2 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/cloud/pcloud_submit.sh b/cloud/pcloud_submit.sh
index a7fb42cb..3c9a1c26 100644
--- a/cloud/pcloud_submit.sh
+++ b/cloud/pcloud_submit.sh
@@ -1,6 +1,6 @@
 TRAIN_MANIFEST="cloud/cloud.manifest.train"
 DEV_MANIFEST="cloud/cloud.manifest.dev"
-CLOUD_MODEL_DIR="/pfs/dlnel/home/USERNAME/deepspeech2/model"
+CLOUD_MODEL_DIR="./checkpoints"
 BATCH_SIZE=256
 NUM_GPU=8
 NUM_NODE=1
@@ -11,7 +11,7 @@ DS2_PATH=${PWD%/*}
 cp -f  pcloud_train.sh ${DS2_PATH}
 
 paddlecloud submit \
--image bootstrapper:5000/wanghaoshuang/pcloud_ds2:latest \
+-image bootstrapper:5000/paddlepaddle/pcloud_ds2:latest \
 -jobname ${JOB_NAME} \
 -cpu ${NUM_GPU} \
 -gpu ${NUM_GPU} \
diff --git a/layer.py b/layer.py
index 3b492645..ef25c0a1 100644
--- a/layer.py
+++ b/layer.py
@@ -55,16 +55,20 @@ def bidirectional_simple_rnn_bn_layer(name, input, size, act):
     :rtype: LayerOutput
     """
     # input-hidden weights shared across bi-direcitonal rnn.
-    input_proj = paddle.layer.fc(
+    input_proj_forward = paddle.layer.fc(
         input=input, size=size, act=paddle.activation.Linear(), bias_attr=False)
-    # batch norm is only performed on input-state projection 
-    input_proj_bn = paddle.layer.batch_norm(
-        input=input_proj, act=paddle.activation.Linear())
+    input_proj_backward = paddle.layer.fc(
+        input=input, size=size, act=paddle.activation.Linear(), bias_attr=False)
+    # batch norm is only performed on input-state projection
+    input_proj_bn_forward = paddle.layer.batch_norm(
+        input=input_proj_forward, act=paddle.activation.Linear())
+    input_proj_bn_backward = paddle.layer.batch_norm(
+        input=input_proj_backward, act=paddle.activation.Linear())
     # forward and backward in time
     forward_simple_rnn = paddle.layer.recurrent(
-        input=input_proj_bn, act=act, reverse=False)
+        input=input_proj_bn_forward, act=act, reverse=False)
     backward_simple_rnn = paddle.layer.recurrent(
-        input=input_proj_bn, act=act, reverse=True)
+        input=input_proj_bn_backward, act=act, reverse=True)
     return paddle.layer.concat(input=[forward_simple_rnn, backward_simple_rnn])