Merge pull request #829 from reyoung/feature/travis_pre_commit_checks

Feature/travis pre commit checks

.travis.yml

@@ -8,10 +8,13 @@ os:
 env:
   - JOB=DOCS
   - JOB=BUILD_AND_TEST
+  - JOB=PRE_COMMIT
 matrix:
   exclude:
     - os: osx
-      env: JOB=DOCS  # Only generate documentation in linux
+      env: JOB=DOCS  # Only generate documentation in linux.
+    - os: osx
+      env: JOB=PRE_COMMIT # Only check pre-commit hook in linux
 
 addons:
   apt:
@@ -39,6 +42,7 @@ addons:
       - lcov
       - graphviz
       - swig
+      - clang-format-3.8
 before_install:
   - |
     if [ ${JOB} == "BUILD_AND_TEST" ]; then
@@ -53,7 +57,8 @@ before_install:
     fi
   - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then sudo paddle/scripts/travis/before_install.linux.sh; fi
   - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then paddle/scripts/travis/before_install.osx.sh; fi
-  - pip install wheel protobuf sphinx recommonmark virtualenv numpy sphinx_rtd_theme
+  - if [[ "$JOB" == "PRE_COMMIT" ]]; then sudo ln -s /usr/bin/clang-format-3.8 /usr/bin/clang-format; fi
+  - pip install wheel protobuf sphinx recommonmark virtualenv numpy sphinx_rtd_theme pre-commit
 script:
   - paddle/scripts/travis/main.sh
 notifications:

WORKSPACE

 # External dependency to Google protobuf.
 http_archive(
-    name = "protobuf",
-    url = "http://github.com/google/protobuf/archive/v3.1.0.tar.gz",
-    sha256 = "0a0ae63cbffc274efb573bdde9a253e3f32e458c41261df51c5dbc5ad541e8f7",
-    strip_prefix = "protobuf-3.1.0",
-)
+    name="protobuf",
+    url="http://github.com/google/protobuf/archive/v3.1.0.tar.gz",
+    sha256="0a0ae63cbffc274efb573bdde9a253e3f32e458c41261df51c5dbc5ad541e8f7",
+    strip_prefix="protobuf-3.1.0", )
 
 # External dependency to gtest 1.7.0.  This method comes from
 # https://www.bazel.io/versions/master/docs/tutorial/cpp.html.
 new_http_archive(
-    name = "gtest",
-    url = "https://github.com/google/googletest/archive/release-1.7.0.zip",
-    sha256 = "b58cb7547a28b2c718d1e38aee18a3659c9e3ff52440297e965f5edffe34b6d0",
-    build_file = "third_party/gtest.BUILD",
-    strip_prefix = "googletest-release-1.7.0",
-)
+    name="gtest",
+    url="https://github.com/google/googletest/archive/release-1.7.0.zip",
+    sha256="b58cb7547a28b2c718d1e38aee18a3659c9e3ff52440297e965f5edffe34b6d0",
+    build_file="third_party/gtest.BUILD",
+    strip_prefix="googletest-release-1.7.0", )

benchmark/tensorflow/rnn/run_multi.sh

@@ -25,4 +25,3 @@ test 4 2 256 512
 test 4 2 512 128 
 test 4 2 512 256 
 test 4 2 512 512 
-

demo/gan/README.md

@@ -10,4 +10,4 @@ Then you can run the command below. The flag -d specifies the training data (cif
 $python gan_trainer.py -d cifar --use_gpu 1
 
 The generated images will be stored in ./cifar_samples/
-The corresponding models will be stored in ./cifar_params/
\ No newline at end of file
+The corresponding models will be stored in ./cifar_params/

demo/gan/data/download_cifar.sh

@@ -15,4 +15,3 @@ set -e
 wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
 tar zxf cifar-10-python.tar.gz
 rm cifar-10-python.tar.gz
-

demo/gan/data/get_mnist_data.sh

@@ -15,5 +15,3 @@ do
         gunzip ${fname}.gz
     fi
 done
-
-

demo/gan/gan_conf.py

@@ -14,10 +14,9 @@
 from paddle.trainer_config_helpers import *
 
 mode = get_config_arg("mode", str, "generator")
-assert mode in set(["generator",
-                    "discriminator",
-                    "generator_training",
-                    "discriminator_training"])
+assert mode in set([
+    "generator", "discriminator", "generator_training", "discriminator_training"
+])
 
 is_generator_training = mode == "generator_training"
 is_discriminator_training = mode == "discriminator_training"
@@ -38,8 +37,8 @@ sample_dim = 2
 settings(
     batch_size=128,
     learning_rate=1e-4,
-    learning_method=AdamOptimizer(beta1=0.5)
-)
+    learning_method=AdamOptimizer(beta1=0.5))
+
 
 def discriminator(sample):
     """
@@ -50,70 +49,87 @@ def discriminator(sample):
     of the sample is from real data.
     """
     param_attr = ParamAttr(is_static=is_generator_training)
-    bias_attr = ParamAttr(is_static=is_generator_training,
-                          initial_mean=1.0,
-                          initial_std=0)
-
-    hidden = fc_layer(input=sample, name="dis_hidden", size=hidden_dim,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=ReluActivation())
-
-    hidden2 = fc_layer(input=hidden, name="dis_hidden2", size=hidden_dim,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=LinearActivation())
-    
-    hidden_bn = batch_norm_layer(hidden2, 
-                     act=ReluActivation(), 
-                     name="dis_hidden_bn", 
-                     bias_attr=bias_attr, 
-                     param_attr=ParamAttr(is_static=is_generator_training,
-                           initial_mean=1.0,
-                           initial_std=0.02),
-                     use_global_stats=False)
-    
-    return fc_layer(input=hidden_bn, name="dis_prob", size=2,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=SoftmaxActivation())
+    bias_attr = ParamAttr(
+        is_static=is_generator_training, initial_mean=1.0, initial_std=0)
+
+    hidden = fc_layer(
+        input=sample,
+        name="dis_hidden",
+        size=hidden_dim,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=ReluActivation())
+
+    hidden2 = fc_layer(
+        input=hidden,
+        name="dis_hidden2",
+        size=hidden_dim,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=LinearActivation())
+
+    hidden_bn = batch_norm_layer(
+        hidden2,
+        act=ReluActivation(),
+        name="dis_hidden_bn",
+        bias_attr=bias_attr,
+        param_attr=ParamAttr(
+            is_static=is_generator_training, initial_mean=1.0,
+            initial_std=0.02),
+        use_global_stats=False)
+
+    return fc_layer(
+        input=hidden_bn,
+        name="dis_prob",
+        size=2,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=SoftmaxActivation())
+
 
 def generator(noise):
     """
     generator generates a sample given noise
     """
     param_attr = ParamAttr(is_static=is_discriminator_training)
-    bias_attr = ParamAttr(is_static=is_discriminator_training,
-                           initial_mean=1.0,
-                           initial_std=0)
-    
-    hidden = fc_layer(input=noise,
-                    name="gen_layer_hidden",
-                    size=hidden_dim,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=ReluActivation())
-
-    hidden2 = fc_layer(input=hidden, name="gen_hidden2", size=hidden_dim,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=LinearActivation())
-    
-    hidden_bn = batch_norm_layer(hidden2, 
-                     act=ReluActivation(), 
-                     name="gen_layer_hidden_bn", 
-                     bias_attr=bias_attr, 
-                     param_attr=ParamAttr(is_static=is_discriminator_training,
-                           initial_mean=1.0,
-                           initial_std=0.02),
-                     use_global_stats=False)
-    
-    return fc_layer(input=hidden_bn,
-                    name="gen_layer1",
-                    size=sample_dim,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=LinearActivation())
+    bias_attr = ParamAttr(
+        is_static=is_discriminator_training, initial_mean=1.0, initial_std=0)
+
+    hidden = fc_layer(
+        input=noise,
+        name="gen_layer_hidden",
+        size=hidden_dim,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=ReluActivation())
+
+    hidden2 = fc_layer(
+        input=hidden,
+        name="gen_hidden2",
+        size=hidden_dim,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=LinearActivation())
+
+    hidden_bn = batch_norm_layer(
+        hidden2,
+        act=ReluActivation(),
+        name="gen_layer_hidden_bn",
+        bias_attr=bias_attr,
+        param_attr=ParamAttr(
+            is_static=is_discriminator_training,
+            initial_mean=1.0,
+            initial_std=0.02),
+        use_global_stats=False)
+
+    return fc_layer(
+        input=hidden_bn,
+        name="gen_layer1",
+        size=sample_dim,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=LinearActivation())
+
 
 if is_generator_training:
     noise = data_layer(name="noise", size=noise_dim)
@@ -126,7 +142,8 @@ if is_generator_training or is_discriminator_training:
     label = data_layer(name="label", size=1)
     prob = discriminator(sample)
     cost = cross_entropy(input=prob, label=label)
-    classification_error_evaluator(input=prob, label=label, name=mode+'_error')
+    classification_error_evaluator(
+        input=prob, label=label, name=mode + '_error')
     outputs(cost)
 
 if is_generator:

demo/gan/gan_conf_image.py

@@ -15,10 +15,9 @@ from paddle.trainer_config_helpers import *
 
 mode = get_config_arg("mode", str, "generator")
 dataSource = get_config_arg("data", str, "mnist")
-assert mode in set(["generator",
-                    "discriminator",
-                    "generator_training",
-                    "discriminator_training"])
+assert mode in set([
+    "generator", "discriminator", "generator_training", "discriminator_training"
+])
 
 is_generator_training = mode == "generator_training"
 is_discriminator_training = mode == "discriminator_training"
@@ -36,24 +35,33 @@ noise_dim = 100
 gf_dim = 64
 df_dim = 64
 if dataSource == "mnist":
-    sample_dim = 28 # image dim
-    c_dim = 1 # image color
+    sample_dim = 28  # image dim
+    c_dim = 1  # image color
 else:
     sample_dim = 32
     c_dim = 3
-s2, s4 = int(sample_dim/2), int(sample_dim/4), 
-s8, s16 = int(sample_dim/8), int(sample_dim/16)
+s2, s4 = int(sample_dim / 2), int(sample_dim / 4),
+s8, s16 = int(sample_dim / 8), int(sample_dim / 16)
 
 settings(
     batch_size=128,
     learning_rate=2e-4,
-    learning_method=AdamOptimizer(beta1=0.5)
-)
+    learning_method=AdamOptimizer(beta1=0.5))
 
-def conv_bn(input, channels, imgSize, num_filters, output_x, stride, name, 
-                 param_attr, bias_attr, param_attr_bn, bn, trans=False, 
-                 act=ReluActivation()):
-    
+
+def conv_bn(input,
+            channels,
+            imgSize,
+            num_filters,
+            output_x,
+            stride,
+            name,
+            param_attr,
+            bias_attr,
+            param_attr_bn,
+            bn,
+            trans=False,
+            act=ReluActivation()):
     """
     conv_bn is a utility function that constructs a convolution/deconv layer 
     with an optional batch_norm layer
@@ -63,10 +71,10 @@ def conv_bn(input, channels, imgSize, num_filters, output_x, stride, name,
     :param trans: whether to use conv (False) or deconv (True)
     :type trans: bool
     """
-    
+
     # calculate the filter_size and padding size based on the given
     # imgSize and ouput size
-    tmp =  imgSize - (output_x - 1) * stride
+    tmp = imgSize - (output_x - 1) * stride
     if tmp <= 1 or tmp > 5:
         raise ValueError("conv input-output dimension does not fit")
     elif tmp <= 3:
@@ -76,111 +84,134 @@ def conv_bn(input, channels, imgSize, num_filters, output_x, stride, name,
         filter_size = tmp
         padding = 0
 
-    print (imgSize, output_x, stride, filter_size, padding)
-    
+    print(imgSize, output_x, stride, filter_size, padding)
+
     if trans:
         nameApx = "_conv"
     else:
         nameApx = "_convt"
-    
+
     if bn:
-        conv = img_conv_layer(input, filter_size=filter_size, 
-                   num_filters=num_filters,
-                   name=name + nameApx, num_channels=channels,
-                   act=LinearActivation(), groups=1, stride=stride, 
-                   padding=padding, bias_attr=bias_attr,
-                   param_attr=param_attr, shared_biases=True, layer_attr=None,
-                   filter_size_y=None, stride_y=None, padding_y=None, 
-                   trans=trans)
-        
-        conv_bn = batch_norm_layer(conv, 
-                         act=act, 
-                         name=name + nameApx + "_bn", 
-                         bias_attr=bias_attr, 
-                         param_attr=param_attr_bn,
-                         use_global_stats=False)
-        
+        conv = img_conv_layer(
+            input,
+            filter_size=filter_size,
+            num_filters=num_filters,
+            name=name + nameApx,
+            num_channels=channels,
+            act=LinearActivation(),
+            groups=1,
+            stride=stride,
+            padding=padding,
+            bias_attr=bias_attr,
+            param_attr=param_attr,
+            shared_biases=True,
+            layer_attr=None,
+            filter_size_y=None,
+            stride_y=None,
+            padding_y=None,
+            trans=trans)
+
+        conv_bn = batch_norm_layer(
+            conv,
+            act=act,
+            name=name + nameApx + "_bn",
+            bias_attr=bias_attr,
+            param_attr=param_attr_bn,
+            use_global_stats=False)
+
         return conv_bn
     else:
-        conv = img_conv_layer(input, filter_size=filter_size, 
-                   num_filters=num_filters,
-                   name=name + nameApx, num_channels=channels,
-                   act=act, groups=1, stride=stride, 
-                   padding=padding, bias_attr=bias_attr,
-                   param_attr=param_attr, shared_biases=True, layer_attr=None,
-                   filter_size_y=None, stride_y=None, padding_y=None,
-                   trans=trans)
+        conv = img_conv_layer(
+            input,
+            filter_size=filter_size,
+            num_filters=num_filters,
+            name=name + nameApx,
+            num_channels=channels,
+            act=act,
+            groups=1,
+            stride=stride,
+            padding=padding,
+            bias_attr=bias_attr,
+            param_attr=param_attr,
+            shared_biases=True,
+            layer_attr=None,
+            filter_size_y=None,
+            stride_y=None,
+            padding_y=None,
+            trans=trans)
         return conv
-    
+
+
 def generator(noise):
     """
     generator generates a sample given noise
     """
-    param_attr = ParamAttr(is_static=is_discriminator_training,
-                           initial_mean=0.0,
-                           initial_std=0.02)
-    bias_attr = ParamAttr(is_static=is_discriminator_training,
-                           initial_mean=0.0,
-                           initial_std=0.0)
-    
-    param_attr_bn=ParamAttr(is_static=is_discriminator_training,
-                           initial_mean=1.0,
-                           initial_std=0.02)
-    
-    h1 = fc_layer(input=noise,
-                    name="gen_layer_h1",
-                    size=s8 * s8 * gf_dim * 4,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=LinearActivation())
-    
-    h1_bn = batch_norm_layer(h1, 
-                     act=ReluActivation(), 
-                     name="gen_layer_h1_bn", 
-                     bias_attr=bias_attr, 
-                     param_attr=param_attr_bn,
-                     use_global_stats=False)
-    
-    h2_bn = conv_bn(h1_bn, 
-                    channels=gf_dim*4, 
-                    output_x=s8,
-                    num_filters=gf_dim*2, 
-                    imgSize=s4,
-                    stride=2,
-                    name="gen_layer_h2", 
-                    param_attr=param_attr, 
-                    bias_attr=bias_attr, 
-                    param_attr_bn=param_attr_bn,
-                    bn=True,
-                    trans=True)
-    
-    h3_bn = conv_bn(h2_bn, 
-                    channels=gf_dim*2, 
-                    output_x=s4,
-                    num_filters=gf_dim, 
-                    imgSize=s2,
-                    stride=2,
-                    name="gen_layer_h3", 
-                    param_attr=param_attr, 
-                    bias_attr=bias_attr, 
-                    param_attr_bn=param_attr_bn,
-                    bn=True,
-                    trans=True)
-     
-    
-    return conv_bn(h3_bn,
-                   channels=gf_dim, 
-                   output_x=s2,
-                   num_filters=c_dim, 
-                   imgSize=sample_dim,
-                   stride=2,
-                   name="gen_layer_h4", 
-                   param_attr=param_attr, 
-                   bias_attr=bias_attr, 
-                   param_attr_bn=param_attr_bn,
-                   bn=False,
-                   trans=True,
-                   act=TanhActivation())
+    param_attr = ParamAttr(
+        is_static=is_discriminator_training, initial_mean=0.0, initial_std=0.02)
+    bias_attr = ParamAttr(
+        is_static=is_discriminator_training, initial_mean=0.0, initial_std=0.0)
+
+    param_attr_bn = ParamAttr(
+        is_static=is_discriminator_training, initial_mean=1.0, initial_std=0.02)
+
+    h1 = fc_layer(
+        input=noise,
+        name="gen_layer_h1",
+        size=s8 * s8 * gf_dim * 4,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=LinearActivation())
+
+    h1_bn = batch_norm_layer(
+        h1,
+        act=ReluActivation(),
+        name="gen_layer_h1_bn",
+        bias_attr=bias_attr,
+        param_attr=param_attr_bn,
+        use_global_stats=False)
+
+    h2_bn = conv_bn(
+        h1_bn,
+        channels=gf_dim * 4,
+        output_x=s8,
+        num_filters=gf_dim * 2,
+        imgSize=s4,
+        stride=2,
+        name="gen_layer_h2",
+        param_attr=param_attr,
+        bias_attr=bias_attr,
+        param_attr_bn=param_attr_bn,
+        bn=True,
+        trans=True)
+
+    h3_bn = conv_bn(
+        h2_bn,
+        channels=gf_dim * 2,
+        output_x=s4,
+        num_filters=gf_dim,
+        imgSize=s2,
+        stride=2,
+        name="gen_layer_h3",
+        param_attr=param_attr,
+        bias_attr=bias_attr,
+        param_attr_bn=param_attr_bn,
+        bn=True,
+        trans=True)
+
+    return conv_bn(
+        h3_bn,
+        channels=gf_dim,
+        output_x=s2,
+        num_filters=c_dim,
+        imgSize=sample_dim,
+        stride=2,
+        name="gen_layer_h4",
+        param_attr=param_attr,
+        bias_attr=bias_attr,
+        param_attr_bn=param_attr_bn,
+        bn=False,
+        trans=True,
+        act=TanhActivation())
 
 
 def discriminator(sample):
@@ -191,58 +222,60 @@ def discriminator(sample):
     of the sample is from generator and dimension 1 is the probabblity
     of the sample is from real data.
     """
-    param_attr = ParamAttr(is_static=is_generator_training,
-                           initial_mean=0.0,
-                           initial_std=0.02)
-    bias_attr = ParamAttr(is_static=is_generator_training,
-                          initial_mean=0.0,
-                          initial_std=0.0)
-    
-    param_attr_bn=ParamAttr(is_static=is_generator_training,
-                           initial_mean=1.0,
-                           initial_std=0.02)
-    
-    h0 = conv_bn(sample, 
-                 channels=c_dim, 
-                 imgSize=sample_dim,
-                 num_filters=df_dim, 
-                 output_x=s2, 
-                 stride=2, 
-                 name="dis_h0", 
-                 param_attr=param_attr, 
-                 bias_attr=bias_attr, 
-                 param_attr_bn=param_attr_bn, 
-                 bn=False)
-    
-    h1_bn = conv_bn(h0, 
-                 channels=df_dim,
-                 imgSize=s2,
-                 num_filters=df_dim*2, 
-                 output_x=s4, 
-                 stride=2, 
-                 name="dis_h1", 
-                 param_attr=param_attr, 
-                 bias_attr=bias_attr, 
-                 param_attr_bn=param_attr_bn, 
-                 bn=True)
-
-    h2_bn = conv_bn(h1_bn, 
-                 channels=df_dim*2,
-                 imgSize=s4,
-                 num_filters=df_dim*4, 
-                 output_x=s8, 
-                 stride=2, 
-                 name="dis_h2", 
-                 param_attr=param_attr, 
-                 bias_attr=bias_attr, 
-                 param_attr_bn=param_attr_bn, 
-                 bn=True)
-        
-    return fc_layer(input=h2_bn, name="dis_prob", size=2,
-                    bias_attr=bias_attr,
-                    param_attr=param_attr,
-                    act=SoftmaxActivation())
+    param_attr = ParamAttr(
+        is_static=is_generator_training, initial_mean=0.0, initial_std=0.02)
+    bias_attr = ParamAttr(
+        is_static=is_generator_training, initial_mean=0.0, initial_std=0.0)
+
+    param_attr_bn = ParamAttr(
+        is_static=is_generator_training, initial_mean=1.0, initial_std=0.02)
+
+    h0 = conv_bn(
+        sample,
+        channels=c_dim,
+        imgSize=sample_dim,
+        num_filters=df_dim,
+        output_x=s2,
+        stride=2,
+        name="dis_h0",
+        param_attr=param_attr,
+        bias_attr=bias_attr,
+        param_attr_bn=param_attr_bn,
+        bn=False)
+
+    h1_bn = conv_bn(
+        h0,
+        channels=df_dim,
+        imgSize=s2,
+        num_filters=df_dim * 2,
+        output_x=s4,
+        stride=2,
+        name="dis_h1",
+        param_attr=param_attr,
+        bias_attr=bias_attr,
+        param_attr_bn=param_attr_bn,
+        bn=True)
+
+    h2_bn = conv_bn(
+        h1_bn,
+        channels=df_dim * 2,
+        imgSize=s4,
+        num_filters=df_dim * 4,
+        output_x=s8,
+        stride=2,
+        name="dis_h2",
+        param_attr=param_attr,
+        bias_attr=bias_attr,
+        param_attr_bn=param_attr_bn,
+        bn=True)
 
+    return fc_layer(
+        input=h2_bn,
+        name="dis_prob",
+        size=2,
+        bias_attr=bias_attr,
+        param_attr=param_attr,
+        act=SoftmaxActivation())
 
 
 if is_generator_training:
@@ -250,13 +283,14 @@ if is_generator_training:
     sample = generator(noise)
 
 if is_discriminator_training:
-    sample = data_layer(name="sample", size=sample_dim * sample_dim*c_dim)
+    sample = data_layer(name="sample", size=sample_dim * sample_dim * c_dim)
 
 if is_generator_training or is_discriminator_training:
     label = data_layer(name="label", size=1)
     prob = discriminator(sample)
     cost = cross_entropy(input=prob, label=label)
-    classification_error_evaluator(input=prob, label=label, name=mode+'_error')
+    classification_error_evaluator(
+        input=prob, label=label, name=mode + '_error')
     outputs(cost)
 
 if is_generator:

demo/gan/gan_trainer.py

@@ -16,7 +16,7 @@ import argparse
 import random
 import numpy
 import cPickle
-import sys,os
+import sys, os
 from PIL import Image
 
 from paddle.trainer.config_parser import parse_config
@@ -24,6 +24,7 @@ from paddle.trainer.config_parser import logger
 import py_paddle.swig_paddle as api
 import matplotlib.pyplot as plt
 
+
 def plot2DScatter(data, outputfile):
     '''
     Plot the data as a 2D scatter plot and save to outputfile
@@ -41,9 +42,11 @@ def plot2DScatter(data, outputfile):
     plt.scatter(x, y)
     plt.savefig(outputfile, bbox_inches='tight')
 
+
 def CHECK_EQ(a, b):
     assert a == b, "a=%s, b=%s" % (a, b)
 
+
 def copy_shared_parameters(src, dst):
     '''
     copy the parameters from src to dst
@@ -52,11 +55,9 @@ def copy_shared_parameters(src, dst):
     :param dst: the destination of the parameters
     :type dst: GradientMachine
     '''
-    src_params = [src.getParameter(i)
-               for i in xrange(src.getParameterSize())]
+    src_params = [src.getParameter(i) for i in xrange(src.getParameterSize())]
     src_params = dict([(p.getName(), p) for p in src_params])
 
-
     for i in xrange(dst.getParameterSize()):
         dst_param = dst.getParameter(i)
         src_param = src_params.get(dst_param.getName(), None)
@@ -67,15 +68,17 @@ def copy_shared_parameters(src, dst):
         CHECK_EQ(len(src_value), len(dst_value))
         dst_value.copyFrom(src_value)
         dst_param.setValueUpdated()
-        
+
+
 def print_parameters(src):
-    src_params = [src.getParameter(i)
-               for i in xrange(src.getParameterSize())]
+    src_params = [src.getParameter(i) for i in xrange(src.getParameterSize())]
 
     print "***************"
     for p in src_params:
         print "Name is %s" % p.getName()
-        print "value is %s \n" % p.getBuf(api.PARAMETER_VALUE).copyToNumpyArray()
+        print "value is %s \n" % p.getBuf(api.PARAMETER_VALUE).copyToNumpyArray(
+        )
+
 
 def load_mnist_data(imageFile):
     f = open(imageFile, "rb")
@@ -86,33 +89,36 @@ def load_mnist_data(imageFile):
         n = 60000
     else:
         n = 10000
-    
-    data = numpy.fromfile(f, 'ubyte', count=n*28*28).reshape((n, 28*28))
+
+    data = numpy.fromfile(f, 'ubyte', count=n * 28 * 28).reshape((n, 28 * 28))
     data = data / 255.0 * 2.0 - 1.0
 
     f.close()
     return data.astype('float32')
 
+
 def load_cifar_data(cifar_path):
     batch_size = 10000
-    data = numpy.zeros((5*batch_size, 32*32*3), dtype = "float32")
+    data = numpy.zeros((5 * batch_size, 32 * 32 * 3), dtype="float32")
     for i in range(1, 6):
         file = cifar_path + "/data_batch_" + str(i)
         fo = open(file, 'rb')
         dict = cPickle.load(fo)
         fo.close()
-        data[(i - 1)*batch_size:(i*batch_size), :] = dict["data"]
-    
+        data[(i - 1) * batch_size:(i * batch_size), :] = dict["data"]
+
     data = data / 255.0 * 2.0 - 1.0
     return data
 
+
 # synthesize 2-D uniform data
 def load_uniform_data():
     data = numpy.random.rand(1000000, 2).astype('float32')
     return data
 
+
 def merge(images, size):
-    if images.shape[1] == 28*28:
+    if images.shape[1] == 28 * 28:
         h, w, c = 28, 28, 1
     else:
         h, w, c = 32, 32, 3
@@ -124,6 +130,7 @@ def merge(images, size):
           ((images[idx, :].reshape((h, w, c), order="F").transpose(1, 0, 2) + 1.0) / 2.0 * 255.0)
     return img.astype('uint8')
 
+
 def save_images(images, path):
     merged_img = merge(images, [8, 8])
     if merged_img.shape[2] == 1:
@@ -131,14 +138,17 @@ def save_images(images, path):
     else:
         im = Image.fromarray(merged_img, mode="RGB")
     im.save(path)
-    
+
+
 def get_real_samples(batch_size, data_np):
-    return data_np[numpy.random.choice(data_np.shape[0], batch_size, 
-                                       replace=False),:]
-    
+    return data_np[numpy.random.choice(
+        data_np.shape[0], batch_size, replace=False), :]
+
+
 def get_noise(batch_size, noise_dim):
     return numpy.random.normal(size=(batch_size, noise_dim)).astype('float32')
 
+
 def get_fake_samples(generator_machine, batch_size, noise):
     gen_inputs = api.Arguments.createArguments(1)
     gen_inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(noise))
@@ -147,12 +157,14 @@ def get_fake_samples(generator_machine, batch_size, noise):
     fake_samples = gen_outputs.getSlotValue(0).copyToNumpyMat()
     return fake_samples
 
+
 def get_training_loss(training_machine, inputs):
     outputs = api.Arguments.createArguments(0)
     training_machine.forward(inputs, outputs, api.PASS_TEST)
     loss = outputs.getSlotValue(0).copyToNumpyMat()
     return numpy.mean(loss)
 
+
 def prepare_discriminator_data_batch_pos(batch_size, data_np):
     real_samples = get_real_samples(batch_size, data_np)
     labels = numpy.ones(batch_size, dtype='int32')
@@ -161,6 +173,7 @@ def prepare_discriminator_data_batch_pos(batch_size, data_np):
     inputs.setSlotIds(1, api.IVector.createVectorFromNumpy(labels))
     return inputs
 
+
 def prepare_discriminator_data_batch_neg(generator_machine, batch_size, noise):
     fake_samples = get_fake_samples(generator_machine, batch_size, noise)
     labels = numpy.zeros(batch_size, dtype='int32')
@@ -169,6 +182,7 @@ def prepare_discriminator_data_batch_neg(generator_machine, batch_size, noise):
     inputs.setSlotIds(1, api.IVector.createVectorFromNumpy(labels))
     return inputs
 
+
 def prepare_generator_data_batch(batch_size, noise):
     label = numpy.ones(batch_size, dtype='int32')
     inputs = api.Arguments.createArguments(2)
@@ -193,10 +207,9 @@ def get_layer_size(model_conf, layer_name):
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("-d", "--data_source", help="mnist or cifar or uniform")
-    parser.add_argument("--use_gpu", default="1", 
-                        help="1 means use gpu for training")
-    parser.add_argument("--gpu_id", default="0", 
-                        help="the gpu_id parameter")
+    parser.add_argument(
+        "--use_gpu", default="1", help="1 means use gpu for training")
+    parser.add_argument("--gpu_id", default="0", help="the gpu_id parameter")
     args = parser.parse_args()
     data_source = args.data_source
     use_gpu = args.use_gpu
@@ -208,30 +221,32 @@ def main():
 
     if not os.path.exists("./%s_params/" % data_source):
         os.makedirs("./%s_params/" % data_source)
-        
-    api.initPaddle('--use_gpu=' + use_gpu, '--dot_period=10', '--log_period=100', 
-                   '--gpu_id=' + args.gpu_id, '--save_dir=' + "./%s_params/" % data_source)
-    
+
+    api.initPaddle('--use_gpu=' + use_gpu, '--dot_period=10',
+                   '--log_period=100', '--gpu_id=' + args.gpu_id,
+                   '--save_dir=' + "./%s_params/" % data_source)
+
     if data_source == "uniform":
         conf = "gan_conf.py"
         num_iter = 10000
     else:
         conf = "gan_conf_image.py"
         num_iter = 1000
-        
+
     gen_conf = parse_config(conf, "mode=generator_training,data=" + data_source)
-    dis_conf = parse_config(conf, "mode=discriminator_training,data=" + data_source)
+    dis_conf = parse_config(conf,
+                            "mode=discriminator_training,data=" + data_source)
     generator_conf = parse_config(conf, "mode=generator,data=" + data_source)
     batch_size = dis_conf.opt_config.batch_size
     noise_dim = get_layer_size(gen_conf.model_config, "noise")
-    
+
     if data_source == "mnist":
         data_np = load_mnist_data("./data/mnist_data/train-images-idx3-ubyte")
     elif data_source == "cifar":
         data_np = load_cifar_data("./data/cifar-10-batches-py/")
     else:
         data_np = load_uniform_data()
-    
+
     # this creates a gradient machine for discriminator
     dis_training_machine = api.GradientMachine.createFromConfigProto(
         dis_conf.model_config)
@@ -244,26 +259,24 @@ def main():
     logger.info(str(generator_conf.model_config))
     generator_machine = api.GradientMachine.createFromConfigProto(
         generator_conf.model_config)
-    
-    dis_trainer = api.Trainer.create(
-        dis_conf, dis_training_machine)
 
-    gen_trainer = api.Trainer.create(
-        gen_conf, gen_training_machine)
-    
+    dis_trainer = api.Trainer.create(dis_conf, dis_training_machine)
+
+    gen_trainer = api.Trainer.create(gen_conf, gen_training_machine)
+
     dis_trainer.startTrain()
     gen_trainer.startTrain()
-    
+
     # Sync parameters between networks (GradientMachine) at the beginning
     copy_shared_parameters(gen_training_machine, dis_training_machine)
     copy_shared_parameters(gen_training_machine, generator_machine)
-    
+
     # constrain that either discriminator or generator can not be trained
     # consecutively more than MAX_strike times
     curr_train = "dis"
     curr_strike = 0
     MAX_strike = 5
-     
+
     for train_pass in xrange(100):
         dis_trainer.startTrainPass()
         gen_trainer.startTrainPass()
@@ -272,23 +285,25 @@ def main():
             noise = get_noise(batch_size, noise_dim)
             data_batch_dis_pos = prepare_discriminator_data_batch_pos(
                 batch_size, data_np)
-            dis_loss_pos = get_training_loss(dis_training_machine, data_batch_dis_pos)
-            
+            dis_loss_pos = get_training_loss(dis_training_machine,
+                                             data_batch_dis_pos)
+
             data_batch_dis_neg = prepare_discriminator_data_batch_neg(
                 generator_machine, batch_size, noise)
-            dis_loss_neg = get_training_loss(dis_training_machine, data_batch_dis_neg)            
-                         
+            dis_loss_neg = get_training_loss(dis_training_machine,
+                                             data_batch_dis_neg)
+
             dis_loss = (dis_loss_pos + dis_loss_neg) / 2.0
-            
+
             # Do forward pass in generator to get the gen_loss
-            data_batch_gen = prepare_generator_data_batch(
-                    batch_size, noise)
+            data_batch_gen = prepare_generator_data_batch(batch_size, noise)
             gen_loss = get_training_loss(gen_training_machine, data_batch_gen)
-             
+
             if i % 100 == 0:
-                print "d_pos_loss is %s     d_neg_loss is %s" % (dis_loss_pos, dis_loss_neg) 
+                print "d_pos_loss is %s     d_neg_loss is %s" % (dis_loss_pos,
+                                                                 dis_loss_neg)
                 print "d_loss is %s    g_loss is %s" % (dis_loss, gen_loss)
-            
+
             # Decide which network to train based on the training history
             # And the relative size of the loss        
             if (not (curr_train == "dis" and curr_strike == MAX_strike)) and \
@@ -297,11 +312,12 @@ def main():
                     curr_strike += 1
                 else:
                     curr_train = "dis"
-                    curr_strike = 1                
+                    curr_strike = 1
                 dis_trainer.trainOneDataBatch(batch_size, data_batch_dis_neg)
-                dis_trainer.trainOneDataBatch(batch_size, data_batch_dis_pos)               
-                copy_shared_parameters(dis_training_machine, gen_training_machine)
- 
+                dis_trainer.trainOneDataBatch(batch_size, data_batch_dis_pos)
+                copy_shared_parameters(dis_training_machine,
+                                       gen_training_machine)
+
             else:
                 if curr_train == "gen":
                     curr_strike += 1
@@ -311,19 +327,23 @@ def main():
                 gen_trainer.trainOneDataBatch(batch_size, data_batch_gen)
                 # TODO: add API for paddle to allow true parameter sharing between different GradientMachines 
                 # so that we do not need to copy shared parameters. 
-                copy_shared_parameters(gen_training_machine, dis_training_machine)
+                copy_shared_parameters(gen_training_machine,
+                                       dis_training_machine)
                 copy_shared_parameters(gen_training_machine, generator_machine)
- 
+
         dis_trainer.finishTrainPass()
         gen_trainer.finishTrainPass()
         # At the end of each pass, save the generated samples/images
         fake_samples = get_fake_samples(generator_machine, batch_size, noise)
         if data_source == "uniform":
-            plot2DScatter(fake_samples, "./%s_samples/train_pass%s.png" % (data_source, train_pass))
+            plot2DScatter(fake_samples, "./%s_samples/train_pass%s.png" %
+                          (data_source, train_pass))
         else:
-            save_images(fake_samples, "./%s_samples/train_pass%s.png" % (data_source, train_pass))
+            save_images(fake_samples, "./%s_samples/train_pass%s.png" %
+                        (data_source, train_pass))
     dis_trainer.finishTrain()
     gen_trainer.finishTrain()
 
+
 if __name__ == '__main__':
     main()

demo/quick_start/trainer_config.resnet-lstm.py

@@ -13,7 +13,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 """
 This configuration is a demonstration of how to implement the stacked LSTM
 with residual connections, i.e. an LSTM layer takes the sum of the hidden states
@@ -46,11 +45,12 @@ is_predict = get_config_arg('is_predict', bool, False)
 trn = 'data/train.list' if not is_predict else None
 tst = 'data/test.list' if not is_predict else 'data/pred.list'
 process = 'process' if not is_predict else 'process_predict'
-define_py_data_sources2(train_list=trn,
-                        test_list=tst,
-                        module="dataprovider_emb",
-                        obj=process,
-                        args={"dictionary": word_dict})
+define_py_data_sources2(
+    train_list=trn,
+    test_list=tst,
+    module="dataprovider_emb",
+    obj=process,
+    args={"dictionary": word_dict})
 
 batch_size = 128 if not is_predict else 1
 settings(
@@ -58,10 +58,9 @@ settings(
     learning_rate=2e-3,
     learning_method=AdamOptimizer(),
     regularization=L2Regularization(8e-4),
-    gradient_clipping_threshold=25
-)
+    gradient_clipping_threshold=25)
 
-bias_attr = ParamAttr(initial_std=0.,l2_rate=0.)
+bias_attr = ParamAttr(initial_std=0., l2_rate=0.)
 
 data = data_layer(name="word", size=len(word_dict))
 emb = embedding_layer(input=data, size=128)
@@ -73,17 +72,15 @@ for i in range(3):
     # The input to the current layer is the sum of the hidden state
     # and input of the previous layer.
     current_input = addto_layer(input=[previous_input, previous_hidden_state])
-    hidden_state = simple_lstm(input=current_input, size=128,
-                               lstm_cell_attr=ExtraAttr(drop_rate=0.1))
+    hidden_state = simple_lstm(
+        input=current_input, size=128, lstm_cell_attr=ExtraAttr(drop_rate=0.1))
     previous_input, previous_hidden_state = current_input, hidden_state
 
 lstm = previous_hidden_state
 
 lstm_last = pooling_layer(input=lstm, pooling_type=MaxPooling())
-output = fc_layer(input=lstm_last, size=2,
-                  bias_attr=bias_attr,
-                  act=SoftmaxActivation())
-
+output = fc_layer(
+    input=lstm_last, size=2, bias_attr=bias_attr, act=SoftmaxActivation())
 
 if is_predict:
     maxid = maxid_layer(output)

demo/semantic_role_labeling/data/extract_dict_feature.py

@@ -33,7 +33,7 @@ def extract_dict_features(pair_file, feature_file):
                 ctx_n1 = sentence_list[verb_index - 1]
             else:
                 ctx_n1 = 'bos'
-            
+
             if verb_index > 1:
                 mark[verb_index - 2] = 1
                 ctx_n2 = sentence_list[verb_index - 2]
@@ -48,7 +48,7 @@ def extract_dict_features(pair_file, feature_file):
                 ctx_p1 = sentence_list[verb_index + 1]
             else:
                 ctx_p1 = 'eos'
-            
+
             if verb_index < len(labels_list) - 3:
                 mark[verb_index + 2] = 1
                 ctx_p2 = sentence_list[verb_index + 2]
@@ -69,7 +69,6 @@ def extract_dict_features(pair_file, feature_file):
             feature_out.write(feature_str + '\n')
 
 
-
 if __name__ == '__main__':
 
     usage = '-p pair_file -f feature_file'

demo/semantic_role_labeling/data/extract_pairs.py

@@ -66,8 +66,8 @@ def transform_labels(sentences, labels):
         else:
             verb_list = []
             for x in labels[i][0]:
-                if x !='-':
-                   verb_list.append(x)
+                if x != '-':
+                    verb_list.append(x)
 
             for j in xrange(1, len(labels[i])):
                 label_list = labels[i][j]
@@ -93,7 +93,7 @@ def transform_labels(sentences, labels):
                         is_in_bracket = True
                     else:
                         print 'error:', ll
-                sen_lab_pair.append((sentences[i], verb_list[j-1], label_seq))
+                sen_lab_pair.append((sentences[i], verb_list[j - 1], label_seq))
     return sen_lab_pair
 
 
@@ -103,7 +103,7 @@ def write_file(sen_lab_pair, output_file):
             sentence = x[0]
             label_seq = ' '.join(x[2])
             assert len(sentence.split()) == len(x[2])
-            fout.write(sentence + '\t' + x[1]+'\t' +label_seq + '\n')
+            fout.write(sentence + '\t' + x[1] + '\t' + label_seq + '\n')
 
 
 if __name__ == '__main__':

demo/semantic_role_labeling/dataprovider.py

@@ -21,7 +21,7 @@ def hook(settings, word_dict, label_dict, predicate_dict, **kwargs):
     settings.word_dict = word_dict
     settings.label_dict = label_dict
     settings.predicate_dict = predicate_dict
-   
+
     #all inputs are integral and sequential type
     settings.slots = [
         integer_value_sequence(len(word_dict)),
@@ -29,25 +29,28 @@ def hook(settings, word_dict, label_dict, predicate_dict, **kwargs):
         integer_value_sequence(len(word_dict)),
         integer_value_sequence(len(word_dict)),
         integer_value_sequence(len(word_dict)),
-        integer_value_sequence(len(word_dict)), 
-        integer_value_sequence(len(predicate_dict)),
-        integer_value_sequence(2),
+        integer_value_sequence(len(word_dict)),
+        integer_value_sequence(len(predicate_dict)), integer_value_sequence(2),
         integer_value_sequence(len(label_dict))
     ]
 
 
 def get_batch_size(yeild_data):
     return len(yeild_data[0])
-    
 
-@provider(init_hook=hook, should_shuffle=True, calc_batch_size=get_batch_size, 
-          can_over_batch_size=False, cache=CacheType.CACHE_PASS_IN_MEM)
+
+@provider(
+    init_hook=hook,
+    should_shuffle=True,
+    calc_batch_size=get_batch_size,
+    can_over_batch_size=False,
+    cache=CacheType.CACHE_PASS_IN_MEM)
 def process(settings, file_name):
     with open(file_name, 'r') as fdata:
         for line in fdata:
             sentence, predicate, ctx_n2, ctx_n1, ctx_0, ctx_p1, ctx_p2,  mark, label = \
                 line.strip().split('\t')
-           
+
             words = sentence.split()
             sen_len = len(words)
             word_slot = [settings.word_dict.get(w, UNK_IDX) for w in words]

demo/semantic_role_labeling/db_lstm.py

@@ -20,7 +20,7 @@ from paddle.trainer_config_helpers import *
 #file paths
 word_dict_file = './data/wordDict.txt'
 label_dict_file = './data/targetDict.txt'
-predicate_file= './data/verbDict.txt'
+predicate_file = './data/verbDict.txt'
 train_list_file = './data/train.list'
 test_list_file = './data/test.list'
 
@@ -47,7 +47,6 @@ if not is_predict:
             w = line.strip()
             predicate_dict[w] = i
 
-
     if is_test:
         train_list_file = None
 
@@ -57,9 +56,11 @@ if not is_predict:
         test_list=test_list_file,
         module='dataprovider',
         obj='process',
-        args={'word_dict': word_dict,
-              'label_dict': label_dict,
-              'predicate_dict': predicate_dict })
+        args={
+            'word_dict': word_dict,
+            'label_dict': label_dict,
+            'predicate_dict': predicate_dict
+        })
 
     word_dict_len = len(word_dict)
     label_dict_len = len(label_dict)
@@ -77,24 +78,16 @@ mark_dim = 5
 hidden_dim = 512
 depth = 8
 
-
-
 ########################### Optimizer #######################################
 
-
 settings(
     batch_size=150,
     learning_method=MomentumOptimizer(momentum=0),
     learning_rate=2e-2,
     regularization=L2Regularization(8e-4),
     is_async=False,
-    model_average=ModelAverage(average_window=0.5,
-                               max_average_window=10000),
-                               
-)
-
-
-
+    model_average=ModelAverage(
+        average_window=0.5, max_average_window=10000), )
 
 ####################################### network ##############################
 #8 features and 1 target
@@ -108,22 +101,28 @@ ctx_p1 = data_layer(name='ctx_p1_data', size=word_dict_len)
 ctx_p2 = data_layer(name='ctx_p2_data', size=word_dict_len)
 mark = data_layer(name='mark_data', size=mark_dict_len)
 
-
 if not is_predict:
     target = data_layer(name='target', size=label_dict_len)
 
-
-default_std=1/math.sqrt(hidden_dim)/3.0
+default_std = 1 / math.sqrt(hidden_dim) / 3.0
 
 emb_para = ParameterAttribute(name='emb', initial_std=0., learning_rate=0.)
 std_0 = ParameterAttribute(initial_std=0.)
-std_default = ParameterAttribute(initial_std=default_std) 
-
-predicate_embedding = embedding_layer(size=word_dim, input=predicate, param_attr=ParameterAttribute(name='vemb',initial_std=default_std))
-mark_embedding = embedding_layer(name='word_ctx-in_embedding', size=mark_dim, input=mark, param_attr=std_0)
-
-word_input=[word, ctx_n2, ctx_n1, ctx_0, ctx_p1, ctx_p2]
-emb_layers = [embedding_layer(size=word_dim, input=x, param_attr=emb_para) for x in word_input]
+std_default = ParameterAttribute(initial_std=default_std)
+
+predicate_embedding = embedding_layer(
+    size=word_dim,
+    input=predicate,
+    param_attr=ParameterAttribute(
+        name='vemb', initial_std=default_std))
+mark_embedding = embedding_layer(
+    name='word_ctx-in_embedding', size=mark_dim, input=mark, param_attr=std_0)
+
+word_input = [word, ctx_n2, ctx_n1, ctx_0, ctx_p1, ctx_p2]
+emb_layers = [
+    embedding_layer(
+        size=word_dim, input=x, param_attr=emb_para) for x in word_input
+]
 emb_layers.append(predicate_embedding)
 emb_layers.append(mark_embedding)
 
@@ -131,84 +130,89 @@ hidden_0 = mixed_layer(
     name='hidden0',
     size=hidden_dim,
     bias_attr=std_default,
-    input=[ full_matrix_projection(input=emb, param_attr=std_default ) for emb in emb_layers ])
-
+    input=[
+        full_matrix_projection(
+            input=emb, param_attr=std_default) for emb in emb_layers
+    ])
 
 mix_hidden_lr = 1e-3
 lstm_para_attr = ParameterAttribute(initial_std=0.0, learning_rate=1.0)
-hidden_para_attr = ParameterAttribute(initial_std=default_std, learning_rate=mix_hidden_lr)
-
-lstm_0 = lstmemory(name='lstm0',
-                   input=hidden_0, 
-                   act=ReluActivation(),
-                   gate_act=SigmoidActivation(),
-                   state_act=SigmoidActivation(),
-                   bias_attr=std_0,
-                   param_attr=lstm_para_attr)
+hidden_para_attr = ParameterAttribute(
+    initial_std=default_std, learning_rate=mix_hidden_lr)
+
+lstm_0 = lstmemory(
+    name='lstm0',
+    input=hidden_0,
+    act=ReluActivation(),
+    gate_act=SigmoidActivation(),
+    state_act=SigmoidActivation(),
+    bias_attr=std_0,
+    param_attr=lstm_para_attr)
 
 #stack L-LSTM and R-LSTM with direct edges
 input_tmp = [hidden_0, lstm_0]
 
-
 for i in range(1, depth):
 
-    mix_hidden = mixed_layer(name='hidden'+str(i),
-                             size=hidden_dim, 
-                             bias_attr=std_default,
-                             input=[full_matrix_projection(input=input_tmp[0], param_attr=hidden_para_attr),
-                                    full_matrix_projection(input=input_tmp[1], param_attr=lstm_para_attr)
-                                   ]
-                             )
-
-    lstm = lstmemory(name='lstm'+str(i),
-                     input=mix_hidden,
-                     act=ReluActivation(),
-                     gate_act=SigmoidActivation(),
-                     state_act=SigmoidActivation(),
-                     reverse=((i % 2)==1),
-                     bias_attr=std_0,
-                     param_attr=lstm_para_attr)
+    mix_hidden = mixed_layer(
+        name='hidden' + str(i),
+        size=hidden_dim,
+        bias_attr=std_default,
+        input=[
+            full_matrix_projection(
+                input=input_tmp[0], param_attr=hidden_para_attr),
+            full_matrix_projection(
+                input=input_tmp[1], param_attr=lstm_para_attr)
+        ])
+
+    lstm = lstmemory(
+        name='lstm' + str(i),
+        input=mix_hidden,
+        act=ReluActivation(),
+        gate_act=SigmoidActivation(),
+        state_act=SigmoidActivation(),
+        reverse=((i % 2) == 1),
+        bias_attr=std_0,
+        param_attr=lstm_para_attr)
 
     input_tmp = [mix_hidden, lstm]
 
-feature_out = mixed_layer(name='output',
-                          size=label_dict_len,
-                          bias_attr=std_default, 
-                          input=[full_matrix_projection(input=input_tmp[0], param_attr=hidden_para_attr),
-                                 full_matrix_projection(input=input_tmp[1], param_attr=lstm_para_attr)
-                                ],
-                          )
-
-
+feature_out = mixed_layer(
+    name='output',
+    size=label_dict_len,
+    bias_attr=std_default,
+    input=[
+        full_matrix_projection(
+            input=input_tmp[0], param_attr=hidden_para_attr),
+        full_matrix_projection(
+            input=input_tmp[1], param_attr=lstm_para_attr)
+    ], )
 
 if not is_predict:
-    crf_l = crf_layer( name = 'crf',
-                       size = label_dict_len,
-                       input = feature_out, 
-                       label = target,
-                       param_attr=ParameterAttribute(name='crfw',initial_std=default_std, learning_rate=mix_hidden_lr)
-
-                      )
-
-    
-    crf_dec_l = crf_decoding_layer(name = 'crf_dec_l',
-                                   size = label_dict_len,
-                                   input = feature_out,
-                                   label = target,
-                                   param_attr=ParameterAttribute(name='crfw')
-                                       )
-
+    crf_l = crf_layer(
+        name='crf',
+        size=label_dict_len,
+        input=feature_out,
+        label=target,
+        param_attr=ParameterAttribute(
+            name='crfw', initial_std=default_std, learning_rate=mix_hidden_lr))
+
+    crf_dec_l = crf_decoding_layer(
+        name='crf_dec_l',
+        size=label_dict_len,
+        input=feature_out,
+        label=target,
+        param_attr=ParameterAttribute(name='crfw'))
 
     eval = sum_evaluator(input=crf_dec_l)
-        
+
     outputs(crf_l)
 
 else:
-    crf_dec_l = crf_decoding_layer(name = 'crf_dec_l',
-                                   size = label_dict_len,
-                                   input = feature_out,
-                                   param_attr=ParameterAttribute(name='crfw')
-                                       )
+    crf_dec_l = crf_decoding_layer(
+        name='crf_dec_l',
+        size=label_dict_len,
+        input=feature_out,
+        param_attr=ParameterAttribute(name='crfw'))
 
     outputs(crf_dec_l)
-

demo/semantic_role_labeling/predict.py

@@ -26,7 +26,8 @@ UNK_IDX = 0
 
 
 class Prediction():
-    def __init__(self, train_conf, dict_file, model_dir, label_file, predicate_dict_file):
+    def __init__(self, train_conf, dict_file, model_dir, label_file,
+                 predicate_dict_file):
         """
         train_conf: trainer configure.
         dict_file: word dictionary file name.
@@ -35,7 +36,7 @@ class Prediction():
 
         self.dict = {}
         self.labels = {}
-        self.predicate_dict={}
+        self.predicate_dict = {}
         self.labels_reverse = {}
         self.load_dict_label(dict_file, label_file, predicate_dict_file)
 
@@ -44,25 +45,18 @@ class Prediction():
         len_pred = len(self.predicate_dict)
 
         conf = parse_config(
-            train_conf,
-            'dict_len=' + str(len_dict) + 
-            ',label_len=' + str(len_label) +
-            ',pred_len=' + str(len_pred) +
-            ',is_predict=True')
+            train_conf, 'dict_len=' + str(len_dict) + ',label_len=' +
+            str(len_label) + ',pred_len=' + str(len_pred) + ',is_predict=True')
         self.network = swig_paddle.GradientMachine.createFromConfigProto(
             conf.model_config)
         self.network.loadParameters(model_dir)
 
         slots = [
-            integer_value_sequence(len_dict),
-            integer_value_sequence(len_dict),
-            integer_value_sequence(len_dict),
-            integer_value_sequence(len_dict),
-            integer_value_sequence(len_dict),
-            integer_value_sequence(len_dict), 
-            integer_value_sequence(len_pred),
-            integer_value_sequence(2)
-            ]
+            integer_value_sequence(len_dict), integer_value_sequence(len_dict),
+            integer_value_sequence(len_dict), integer_value_sequence(len_dict),
+            integer_value_sequence(len_dict), integer_value_sequence(len_dict),
+            integer_value_sequence(len_pred), integer_value_sequence(2)
+        ]
         self.converter = DataProviderConverter(slots)
 
     def load_dict_label(self, dict_file, label_file, predicate_dict_file):
@@ -78,6 +72,7 @@ class Prediction():
 
         for line_count, line in enumerate(open(predicate_dict_file, 'r')):
             self.predicate_dict[line.strip()] = line_count
+
     def get_data(self, data_file):
         """
         Get input data of paddle format.
@@ -88,9 +83,10 @@ class Prediction():
                 ).split('\t')
                 words = sentence.split()
                 sen_len = len(words)
-                 
+
                 word_slot = [self.dict.get(w, UNK_IDX) for w in words]
-                predicate_slot = [self.predicate_dict.get(predicate, UNK_IDX)] * sen_len
+                predicate_slot = [self.predicate_dict.get(predicate, UNK_IDX)
+                                  ] * sen_len
                 ctx_n2_slot = [self.dict.get(ctx_n2, UNK_IDX)] * sen_len
                 ctx_n1_slot = [self.dict.get(ctx_n1, UNK_IDX)] * sen_len
                 ctx_0_slot = [self.dict.get(ctx_0, UNK_IDX)] * sen_len
@@ -99,7 +95,7 @@ class Prediction():
 
                 marks = mark.split()
                 mark_slot = [int(w) for w in marks]
-                
+
                 yield word_slot, ctx_n2_slot, ctx_n1_slot, \
                       ctx_0_slot, ctx_p1_slot, ctx_p2_slot, predicate_slot, mark_slot
 
@@ -123,8 +119,9 @@ class Prediction():
 
 
 def option_parser():
-    usage = ("python predict.py -c config -w model_dir " 
-             "-d word dictionary -l label_file -i input_file  -p pred_dict_file")
+    usage = (
+        "python predict.py -c config -w model_dir "
+        "-d word dictionary -l label_file -i input_file  -p pred_dict_file")
     parser = OptionParser(usage="usage: %s [options]" % usage)
     parser.add_option(
         "-c",
@@ -187,8 +184,9 @@ def main():
     output_file = options.output_file
 
     swig_paddle.initPaddle("--use_gpu=0")
-    predict = Prediction(train_conf, dict_file, model_path, label_file, predict_dict_file)
-    predict.predict(data_file,output_file)
+    predict = Prediction(train_conf, dict_file, model_path, label_file,
+                         predict_dict_file)
+    predict.predict(data_file, output_file)
 
 
 if __name__ == '__main__':

demo/sentiment/predict.py

@@ -71,9 +71,7 @@ class SentimentPrediction():
         transform word into integer index according to the dictionary.
         """
         words = data.strip().split()
-        word_slot = [
-            self.word_dict[w] for w in words if w in self.word_dict
-        ]
+        word_slot = [self.word_dict[w] for w in words if w in self.word_dict]
         return word_slot
 
     def batch_predict(self, data_batch):
@@ -85,8 +83,8 @@ class SentimentPrediction():
             if self.label is None:
                 print("predicting label is %d" % (lab[0]))
             else:
-                print("predicting label is %s" %
-                      (self.label[lab[0]]))
+                print("predicting label is %s" % (self.label[lab[0]]))
+
 
 def option_parser():
     usage = "python predict.py -n config -w model_dir -d dictionary -i input_file "
@@ -143,9 +141,10 @@ def main():
         batch.append([predict.get_index(line)])
         if len(batch) == batch_size:
             predict.batch_predict(batch)
-            batch=[]
+            batch = []
     if len(batch) > 0:
         predict.batch_predict(batch)
 
+
 if __name__ == '__main__':
     main()

doc_cn/cluster/k8s/distributed_training_on_kubernetes.md

@@ -306,4 +306,4 @@ I1116 09:10:18.019069    50 ParameterClient2.cpp:122] pserver 2 192.168.223.143:
 I1116 09:10:18.019492    50 ParameterClient2.cpp:122] pserver 3 192.168.223.143:7165
 I1116 09:10:18.019716    50 ParameterClient2.cpp:122] pserver 4 192.168.129.71:7164
 I1116 09:10:18.019836    50 ParameterClient2.cpp:122] pserver 5 192.168.129.71:7165
-```
\ No newline at end of file
+```

doc_cn/cluster/k8s/job.yaml

@@ -40,4 +40,4 @@ spec:
         - name: jobpath
           mountPath: /home/jobpath       
       restartPolicy: Never
-    
\ No newline at end of file
+    

doc_cn/cluster/k8s/start_paddle.py

@@ -19,7 +19,6 @@ import socket
 import os
 import argparse
 
-
 # configuration for cluster
 API = "/api/v1/namespaces/"
 JOBSELECTOR = "labelSelector=job-name="
@@ -145,8 +144,8 @@ def startPaddle(idMap={}, train_args_dict=None):
 
 
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser(prog="start_paddle.py",
-                                     description='simple tool for k8s')
+    parser = argparse.ArgumentParser(
+        prog="start_paddle.py", description='simple tool for k8s')
     args, train_args_list = parser.parse_known_args()
     train_args = refine_unknown_args(train_args_list)
     train_args_dict = dict(zip(train_args[:-1:2], train_args[1::2]))

doc_cn/demo/sentiment_analysis/index.rst

-情感分析教程
-===========================
-
-.. toctree::
-    :maxdepth: 3
-    :glob:
-
+情感分析教程
+===========================
+
+.. toctree::
+    :maxdepth: 3
+    :glob:
+
     Training Locally <sentiment_analysis.md>
\ No newline at end of file

doc_theme/static/js/paddle_doc_init.js

@@ -28,4 +28,4 @@ $(document).ready(function(){
     $('.doc-menu-vertical').find('li.current').last().addClass('active');
 
     $('.doc-menu-vertical').perfectScrollbar();
-});
\ No newline at end of file
+});

paddle/api/GradientMachine.cpp

@@ -15,8 +15,8 @@ limitations under the License. */
 #include "PaddleAPI.h"
 #include "PaddleAPIPrivate.h"
 
-#include "paddle/gserver/gradientmachines/NeuralNetwork.h"
 #include "Internal.h"
+#include "paddle/gserver/gradientmachines/NeuralNetwork.h"
 
 std::vector<int> GradientMachine::defaultParamTypes = {
     PARAMETER_VALUE, PARAMETER_GRADIENT, PARAMETER_MOMENTUM};

paddle/api/Internal.h

@@ -16,14 +16,13 @@ limitations under the License. */
 
 #include "PaddleAPI.h"
 
-#include <vector>
 #include <algorithm>
+#include <vector>
 
 template <typename T1, typename T2>
 void staticCastVector(std::vector<T2>* dest, const std::vector<T1>& src) {
   dest->resize(src.size());
-  std::transform(src.begin(),
-                 src.end(),
-                 dest->begin(),
-                 [](T1 t) { return static_cast<T2>(t); });
+  std::transform(src.begin(), src.end(), dest->begin(), [](T1 t) {
+    return static_cast<T2>(t);
+  });
 }

paddle/api/Matrix.cpp

@@ -12,12 +12,12 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "PaddleAPI.h"
 #include "paddle/math/Matrix.h"
-#include "paddle/math/SparseMatrix.h"
-#include "paddle/math/CpuSparseMatrix.h"
-#include <iostream>
 #include <cstring>
+#include <iostream>
+#include "PaddleAPI.h"
+#include "paddle/math/CpuSparseMatrix.h"
+#include "paddle/math/SparseMatrix.h"
 
 struct MatrixPrivate {
   std::shared_ptr<paddle::Matrix> mat;

paddle/api/PaddleAPI.h

@@ -16,8 +16,8 @@ limitations under the License. */
 
 #include <stddef.h>
 #include <stdint.h>
-#include <string>
 #include <stdexcept>
+#include <string>
 #include <vector>
 #include "paddle/utils/GlobalConstants.h"
 #include "paddle/utils/TypeDefs.h"

paddle/api/Parameter.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "PaddleAPI.h"
 #include "paddle/parameter/Parameter.h"
+#include "PaddleAPI.h"
 
 struct ParameterPrivate {
   std::shared_ptr<paddle::Parameter> sharedPtr;

paddle/api/ParameterOptimizer.cpp

@@ -12,11 +12,11 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "PaddleAPI.h"
-#include "PaddleAPIPrivate.h"
 #include "paddle/parameter/ParameterOptimizer.h"
-#include "Internal.h"
 #include <algorithm>
+#include "Internal.h"
+#include "PaddleAPI.h"
+#include "PaddleAPIPrivate.h"
 
 struct ParameterOptimizerPrivate {
   std::unique_ptr<paddle::ParameterOptimizer> optimizer;
@@ -36,16 +36,13 @@ struct ParameterTraverseCallbackPrivate {
              size_t sparseId) {
     std::vector<paddle::VectorPtr> real_vecs;
     real_vecs.resize(vecs.size());
-    std::transform(vecs.begin(),
-                   vecs.end(),
-                   real_vecs.begin(),
-                   [](Vector* v) {
-                     if (v) {
-                       return *(paddle::VectorPtr*)(v->getSharedPtr());
-                     } else {
-                       return paddle::VectorPtr();
-                     }
-                   });
+    std::transform(vecs.begin(), vecs.end(), real_vecs.begin(), [](Vector* v) {
+      if (v) {
+        return *(paddle::VectorPtr*)(v->getSharedPtr());
+      } else {
+        return paddle::VectorPtr();
+      }
+    });
 
     paddle::ParameterConfig& real_conf =
         *(paddle::ParameterConfig*)(const_cast<ParameterConfig&>(conf)

paddle/api/SequenceGenerator.cpp

@@ -12,14 +12,14 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include <algorithm>
+#include <iterator>
+#include <sstream>
+#include <vector>
 #include "PaddleAPI.h"
 #include "paddle/gserver/gradientmachines/GradientMachine.h"
 #include "paddle/parameter/Argument.h"
 #include "paddle/utils/Flags.h"
-#include <vector>
-#include <sstream>
-#include <algorithm>
-#include <iterator>
 
 // used to represent partial sequence
 struct Path {

paddle/api/Trainer.cpp

@@ -16,12 +16,12 @@ limitations under the License. */
 #include "PaddleAPIPrivate.h"
 
 #include <stdlib.h>
-#include <memory>
 #include <atomic>
+#include <memory>
 
+#include "paddle/gserver/gradientmachines/NeuralNetwork.h"
 #include "paddle/trainer/ParamUtil.h"
 #include "paddle/trainer/Trainer.h"
-#include "paddle/gserver/gradientmachines/NeuralNetwork.h"
 #include "paddle/trainer/TrainerInternal.h"
 #include "paddle/utils/Flags.h"
 

paddle/api/Util.cpp

@@ -14,16 +14,16 @@ limitations under the License. */
 
 #include "PaddleAPI.h"
 
-#include "paddle/utils/Util.h"
-#include "paddle/utils/PythonUtil.h"
-#include "paddle/utils/Flags.h"
-#include "paddle/utils/Excepts.h"
 #include "paddle/parameter/Parameter.h"
+#include "paddle/utils/Excepts.h"
+#include "paddle/utils/Flags.h"
+#include "paddle/utils/PythonUtil.h"
+#include "paddle/utils/Util.h"
 
 #include <fenv.h>
+#include <algorithm>
 #include <iostream>
 #include <iterator>
-#include <algorithm>
 
 void initPaddle(int argc, char** argv) {
   paddle::initMain(argc, argv);

paddle/api/Vector.cpp

@@ -282,7 +282,7 @@ FloatArray Vector::getData() const {
 }
 
 void Vector::copyFrom(Vector* src) throw(RangeError) {
-  if (src->m->vec->getSize() !=  m->vec->getSize()) {
+  if (src->m->vec->getSize() != m->vec->getSize()) {
     throw RangeError();
   }
   m->vec->copyFrom(*src->m->vec);

paddle/api/test/testMatrix.py

@@ -100,11 +100,12 @@ class TestMatrix(unittest.TestCase):
 
             for a, e in zip(gpu_m.getData(), [1.0, 3.23, 3.0, 4.0, 5.0, 6.0]):
                 self.assertAlmostEqual(a, e)
-    
+
     def test_numpy(self):
         numpy_mat = np.matrix([[1, 2], [3, 4], [5, 6]], dtype="float32")
         m = swig_paddle.Matrix.createDenseFromNumpy(numpy_mat)
-        self.assertEqual((int(m.getHeight()), int(m.getWidth())), numpy_mat.shape)
+        self.assertEqual((int(m.getHeight()), int(m.getWidth())),
+                         numpy_mat.shape)
         self.assertEqual(m.isGpu(), swig_paddle.isUsingGpu())
         for a, e in zip(m.getData(), [1.0, 2.0, 3.0, 4.0, 5.0, 6.0]):
             self.assertAlmostEqual(a, e)

paddle/api/test/testVector.py

@@ -26,17 +26,17 @@ class TestIVector(unittest.TestCase):
             self.assertEqual(m[i], 0)
             m[i] = i
             self.assertEqual(m[i], i)
-        
+
         m = swig_paddle.IVector.createZero(10)
         self.assertEqual(m.isGpu(), swig_paddle.isUsingGpu())
-        self.assertEqual(m.getData(), [0]*10)
+        self.assertEqual(m.getData(), [0] * 10)
 
     def test_create(self):
         m = swig_paddle.IVector.create(range(10), False)
         self.assertIsNotNone(m)
         for i in xrange(10):
             self.assertEqual(m[i], i)
-        
+
         m = swig_paddle.IVector.create(range(10))
         self.assertEqual(m.isGpu(), swig_paddle.isUsingGpu())
         self.assertEqual(m.getData(), range(10))
@@ -69,7 +69,7 @@ class TestIVector(unittest.TestCase):
             expect_vec = range(0, 10)
             expect_vec[4] = 7
             self.assertEqual(vec.getData(), expect_vec)
-    
+
     def test_numpy(self):
         vec = np.array([1, 3, 4, 65, 78, 1, 4], dtype="int32")
         iv = swig_paddle.IVector.createVectorFromNumpy(vec)
@@ -85,10 +85,10 @@ class TestVector(unittest.TestCase):
             self.assertTrue(util.doubleEqual(v[i], 0))
             v[i] = i
             self.assertTrue(util.doubleEqual(v[i], i))
-        
+
         v = swig_paddle.Vector.createZero(10)
         self.assertEqual(v.isGpu(), swig_paddle.isUsingGpu())
-        self.assertEqual(v.getData(), [0]*10)
+        self.assertEqual(v.getData(), [0] * 10)
 
     def testCreate(self):
         v = swig_paddle.Vector.create([x / 100.0 for x in xrange(100)], False)
@@ -96,14 +96,13 @@ class TestVector(unittest.TestCase):
         for i in xrange(len(v)):
             self.assertTrue(util.doubleEqual(v[i], i / 100.0))
         self.assertEqual(100, len(v))
-        
+
         v = swig_paddle.Vector.create([x / 100.0 for x in xrange(100)])
         self.assertEqual(v.isGpu(), swig_paddle.isUsingGpu())
         self.assertEqual(100, len(v))
         vdata = v.getData()
         for i in xrange(len(v)):
             self.assertTrue(util.doubleEqual(vdata[i], i / 100.0))
-        
 
     def testCpuNumpy(self):
         numpy_arr = np.array([1.2, 2.3, 3.4, 4.5], dtype="float32")
@@ -128,7 +127,7 @@ class TestVector(unittest.TestCase):
 
         for i in xrange(1, len(numpy_3)):
             util.doubleEqual(numpy_3[i], vec[i])
-    
+
     def testNumpy(self):
         numpy_arr = np.array([1.2, 2.3, 3.4, 4.5], dtype="float32")
         vec = swig_paddle.Vector.createVectorFromNumpy(numpy_arr)
@@ -136,7 +135,6 @@ class TestVector(unittest.TestCase):
         vecData = vec.getData()
         for n, v in zip(numpy_arr, vecData):
             self.assertTrue(util.doubleEqual(n, v))
-        
 
     def testCopyFromNumpy(self):
         vec = swig_paddle.Vector.createZero(1, False)

paddle/cuda/include/hl_base.h

@@ -223,9 +223,9 @@ typedef struct {
 
 #ifdef __NVCC__
 
-#include "paddle/utils/Logging.h"
-#include "hl_cuda.h"
 #include "cuda_runtime.h"
+#include "hl_cuda.h"
+#include "paddle/utils/Logging.h"
 
 extern __thread bool g_sync_flag;
 extern __thread cudaStream_t default_stream;

paddle/cuda/include/hl_dso_loader.h

@@ -16,8 +16,8 @@ limitations under the License. */
 #define HL_DSO_LOADER_H_
 
 #include <dlfcn.h>
-#include <string>
 #include <memory>
+#include <string>
 #include "hl_base.h"
 
 /**

paddle/cuda/include/hl_gpu.h

@@ -15,28 +15,28 @@ limitations under the License. */
 #ifndef HL_GPU_H_
 #define HL_GPU_H_
 
+#include "hl_aggregate.h"
 #include "hl_base.h"
+#include "hl_cnn.h"
 #include "hl_cuda.h"
 #include "hl_cuda_cublas.h"
 #include "hl_cuda_cudnn.h"
-#include "hl_matrix.h"
-#include "hl_aggregate.h"
-#include "hl_cnn.h"
-#include "hl_sparse.h"
 #include "hl_lstm.h"
+#include "hl_matrix.h"
 #include "hl_sequence.h"
+#include "hl_sparse.h"
 #include "hl_warpctc_wrap.h"
 
 #ifdef HPPL_STUB_FUNC
-#include "stub/hl_cuda_stub.h"
-#include "stub/hl_cuda_cublas_stub.h"
-#include "stub/hl_cuda_cudnn_stub.h"
-#include "stub/hl_matrix_stub.h"
 #include "stub/hl_aggregate_stub.h"
 #include "stub/hl_cnn_stub.h"
-#include "stub/hl_sparse_stub.h"
+#include "stub/hl_cuda_cublas_stub.h"
+#include "stub/hl_cuda_cudnn_stub.h"
+#include "stub/hl_cuda_stub.h"
 #include "stub/hl_lstm_stub.h"
+#include "stub/hl_matrix_stub.h"
 #include "stub/hl_sequence_stub.h"
+#include "stub/hl_sparse_stub.h"
 #endif
 
 #endif /* HL_GPU_H_ */

paddle/cuda/include/hl_time.h

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #ifndef HL_TIME_H_
 #define HL_TIME_H_
-
+#include <cstdint>
 /**
  * @brief   High resolution timer.
  *

paddle/cuda/src/hl_cuda_cublas.cc

@@ -12,12 +12,12 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "hl_cuda_cublas.h"
 #include <sys/time.h>
 #include <mutex>
 #include "hl_cuda.h"
-#include "hl_cuda_cublas.h"
-#include "hl_thread.ph"
 #include "hl_dso_loader.h"
+#include "hl_thread.ph"
 #include "paddle/utils/Logging.h"
 
 namespace dynload {

paddle/cuda/src/hl_cuda_cudnn.cc

@@ -12,14 +12,14 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "hl_cuda_cudnn.h"
 #include <cudnn.h>
 #include <mutex>
-#include "hl_cuda_cudnn.h"
 #include "hl_cuda_cudnn.ph"
-#include "hl_thread.ph"
 #include "hl_dso_loader.h"
-#include "paddle/utils/Logging.h"
+#include "hl_thread.ph"
 #include "paddle/utils/CommandLineParser.h"
+#include "paddle/utils/Logging.h"
 
 P_DEFINE_int32(cudnn_conv_workspace_limit_in_mb,
                4096,

paddle/cuda/src/hl_cuda_device.cc

@@ -12,13 +12,13 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "hl_cuda.h"
 #include <cuda_profiler_api.h>
 #include <string.h>
 #include <sys/syscall.h>
 #include <sys/time.h>
 #include <unistd.h>
 #include <mutex>
-#include "hl_cuda.h"
 #include "hl_cuda.ph"
 #include "hl_dso_loader.h"
 #include "hl_thread.ph"

paddle/cuda/src/hl_cudart_wrap.cc

@@ -14,8 +14,8 @@ limitations under the License. */
 
 #ifdef PADDLE_USE_DSO
 
-#include <mutex>
 #include <cuda_runtime.h>
+#include <mutex>
 #include "hl_dso_loader.h"
 
 /**

paddle/cuda/src/hl_time.cc

@@ -12,10 +12,11 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <chrono>
+#include "hl_time.h"
 #include <stdlib.h>
+#include <chrono>
+#include <cstdint>
 #include <iostream>
-#include "hl_time.h"
 
 using std::chrono::high_resolution_clock;
 

paddle/cuda/src/hl_warpctc_wrap.cc

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <mutex>
 #include "hl_warpctc_wrap.h"
+#include <mutex>
 #include "hl_dso_loader.h"
 #include "paddle/utils/Logging.h"
 

paddle/gserver/activations/ActivationFunction.cpp

@@ -15,13 +15,13 @@ limitations under the License. */
 #include "ActivationFunction.h"
 
 #include <algorithm>
-#include <memory>
 #include <iostream>
-#include <type_traits>
+#include <memory>
 #include <string>
 #include <thread>
-#include "paddle/utils/ClassRegistrar.h"
+#include <type_traits>
 #include "paddle/parameter/Argument.h"
+#include "paddle/utils/ClassRegistrar.h"
 
 #include "paddle/utils/Logging.h"
 

paddle/gserver/dataproviders/DataProvider.cpp

@@ -14,12 +14,12 @@ limitations under the License. */
 
 #include "DataProvider.h"
 
-#include "paddle/utils/Util.h"
-#include "paddle/utils/StringUtil.h"
-#include "paddle/utils/Logging.h"
-#include <algorithm>
 #include <unistd.h>
+#include <algorithm>
 #include "ProtoDataProvider.h"
+#include "paddle/utils/Logging.h"
+#include "paddle/utils/StringUtil.h"
+#include "paddle/utils/Util.h"
 
 namespace paddle {
 

paddle/gserver/dataproviders/DataProvider.h

@@ -14,28 +14,28 @@ limitations under the License. */
 
 #pragma once
 
-#include <vector>
-#include <memory>
-#include <mutex>
-#include <iostream>
-#include <fstream>
 #include <stdint.h>
-#include <string.h>
-#include <stdlib.h>
 #include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <fstream>
+#include <iostream>
+#include <memory>
+#include <mutex>
+#include <vector>
 
+#include "DataConfig.pb.h"
+#include "paddle/math/Matrix.h"
+#include "paddle/math/SparseMatrix.h"
+#include "paddle/math/Vector.h"
+#include "paddle/parameter/Argument.h"
+#include "paddle/utils/ClassRegistrar.h"
+#include "paddle/utils/Locks.h"
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Queue.h"
-#include "paddle/utils/Locks.h"
 #include "paddle/utils/ThreadLocal.h"
 #include "paddle/utils/TypeDefs.h"
-#include "paddle/math/Matrix.h"
-#include "paddle/math/SparseMatrix.h"
 #include "paddle/utils/Util.h"
-#include "paddle/math/Vector.h"
-#include "DataConfig.pb.h"
-#include "paddle/utils/ClassRegistrar.h"
-#include "paddle/parameter/Argument.h"
 
 namespace paddle {
 /**

paddle/gserver/dataproviders/MultiDataProvider.cpp

@@ -12,10 +12,10 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Util.h"
 #include "MultiDataProvider.h"
-#include "paddle/utils/Logging.h"
 #include <algorithm>
+#include "paddle/utils/Logging.h"
+#include "paddle/utils/Util.h"
 
 namespace paddle {
 

paddle/gserver/dataproviders/ProtoDataProvider.cpp

@@ -13,14 +13,14 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "ProtoDataProvider.h"
-#include "paddle/utils/Util.h"
-#include "paddle/utils/StringUtil.h"
 #include <algorithm>
 #include <fstream>
 #include <istream>
+#include "paddle/utils/StringUtil.h"
+#include "paddle/utils/Util.h"
 
-#include "paddle/utils/Logging.h"
 #include "DataProviderGroup.h"
+#include "paddle/utils/Logging.h"
 
 P_DEFINE_double(memory_threshold_on_load_data,
                 1.0,
@@ -562,16 +562,16 @@ int64_t ProtoDataProvider::getNextBatchInternal(int64_t size,
         auto mat = cpuArguments[slot].value;
         mat->resize(size, dim);
         if (std::dynamic_pointer_cast<GpuSparseMatrix>(mat)) {
-          std::dynamic_pointer_cast<GpuSparseMatrix>(mat)
-              ->copyFrom(dataPos.data(),
-                         slots_[slot].indices.data(),
-                         slots_[slot].sparseNonValueData.data(),
-                         HPPL_STREAM_1);
+          std::dynamic_pointer_cast<GpuSparseMatrix>(mat)->copyFrom(
+              dataPos.data(),
+              slots_[slot].indices.data(),
+              slots_[slot].sparseNonValueData.data(),
+              HPPL_STREAM_1);
         } else if (std::dynamic_pointer_cast<CpuSparseMatrix>(mat)) {
-          std::dynamic_pointer_cast<CpuSparseMatrix>(mat)
-              ->copyFrom(dataPos.data(),
-                         slots_[slot].indices.data(),
-                         slots_[slot].sparseNonValueData.data());
+          std::dynamic_pointer_cast<CpuSparseMatrix>(mat)->copyFrom(
+              dataPos.data(),
+              slots_[slot].indices.data(),
+              slots_[slot].sparseNonValueData.data());
         } else {
           LOG(FATAL) << "Not Supported";
         }
@@ -598,16 +598,16 @@ int64_t ProtoDataProvider::getNextBatchInternal(int64_t size,
         auto mat = cpuArguments[slot].value;
         mat->resize(size, dim);
         if (std::dynamic_pointer_cast<GpuSparseMatrix>(mat)) {
-          std::dynamic_pointer_cast<GpuSparseMatrix>(mat)
-              ->copyFrom(dataPos.data(),
-                         slots_[slot].indices.data(),
-                         slots_[slot].sparseFloatValueData.data(),
-                         HPPL_STREAM_1);
+          std::dynamic_pointer_cast<GpuSparseMatrix>(mat)->copyFrom(
+              dataPos.data(),
+              slots_[slot].indices.data(),
+              slots_[slot].sparseFloatValueData.data(),
+              HPPL_STREAM_1);
         } else if (std::dynamic_pointer_cast<CpuSparseMatrix>(mat)) {
-          std::dynamic_pointer_cast<CpuSparseMatrix>(mat)
-              ->copyFrom(dataPos.data(),
-                         slots_[slot].indices.data(),
-                         slots_[slot].sparseFloatValueData.data());
+          std::dynamic_pointer_cast<CpuSparseMatrix>(mat)->copyFrom(
+              dataPos.data(),
+              slots_[slot].indices.data(),
+              slots_[slot].sparseFloatValueData.data());
         } else {
           LOG(FATAL) << "Not Supported";
         }

paddle/gserver/dataproviders/ProtoDataProvider.h

@@ -16,8 +16,8 @@ limitations under the License. */
 
 #include <vector>
 
-#include "paddle/utils/Stat.h"
 #include "DataFormat.pb.h"
+#include "paddle/utils/Stat.h"
 
 #include "DataProvider.h"
 #include "ProtoReader.h"

paddle/gserver/dataproviders/ProtoReader.h

@@ -16,10 +16,10 @@ limitations under the License. */
 
 #include <memory>
 
-#include <google/protobuf/message_lite.h>
 #include <google/protobuf/io/coded_stream.h>
-#include <google/protobuf/io/zero_copy_stream_impl.h>
 #include <google/protobuf/io/gzip_stream.h>
+#include <google/protobuf/io/zero_copy_stream_impl.h>
+#include <google/protobuf/message_lite.h>
 
 namespace paddle {
 

paddle/gserver/dataproviders/PyDataProvider.cpp

@@ -13,10 +13,10 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "PyDataProvider.h"
-#include "paddle/utils/PythonUtil.h"
 #include <fenv.h>
-#include "paddle/utils/Util.h"
 #include "paddle/utils/Excepts.h"
+#include "paddle/utils/PythonUtil.h"
+#include "paddle/utils/Util.h"
 
 namespace paddle {
 
@@ -316,16 +316,16 @@ void PyDataProvider::handleSparseNonValueSlot(
   auto mat = cpuArguments[slotIndex].value;
   mat->resize(slot.sampleNum, dim, slot.sampleNum, NO_VALUE, SPARSE_CSR);
   if (std::dynamic_pointer_cast<GpuSparseMatrix>(mat)) {
-    std::dynamic_pointer_cast<GpuSparseMatrix>(mat)
-        ->copyFrom(slot.sampleSequenceIdVec.data(),
-                   slot.indices.data(),
-                   slot.sparseNonValueData.data(),
-                   HPPL_STREAM_1);
+    std::dynamic_pointer_cast<GpuSparseMatrix>(mat)->copyFrom(
+        slot.sampleSequenceIdVec.data(),
+        slot.indices.data(),
+        slot.sparseNonValueData.data(),
+        HPPL_STREAM_1);
   } else if (std::dynamic_pointer_cast<CpuSparseMatrix>(mat)) {
-    std::dynamic_pointer_cast<CpuSparseMatrix>(mat)
-        ->copyFrom(slot.sampleSequenceIdVec.data(),
-                   slot.indices.data(),
-                   slot.sparseNonValueData.data());
+    std::dynamic_pointer_cast<CpuSparseMatrix>(mat)->copyFrom(
+        slot.sampleSequenceIdVec.data(),
+        slot.indices.data(),
+        slot.sparseNonValueData.data());
   } else {
     LOG(FATAL) << "Not Supported";
   }
@@ -347,16 +347,16 @@ void PyDataProvider::handleSparseValueSlot(
   auto mat = cpuArguments[slotIndex].value;
   mat->resize(slot.sampleNum, dim, slot.sampleNum, FLOAT_VALUE, SPARSE_CSR);
   if (std::dynamic_pointer_cast<GpuSparseMatrix>(mat)) {
-    std::dynamic_pointer_cast<GpuSparseMatrix>(mat)
-        ->copyFrom(slot.sampleSequenceIdVec.data(),
-                   slot.indices.data(),
-                   slot.sparseFloatValueData.data(),
-                   HPPL_STREAM_DEFAULT);
+    std::dynamic_pointer_cast<GpuSparseMatrix>(mat)->copyFrom(
+        slot.sampleSequenceIdVec.data(),
+        slot.indices.data(),
+        slot.sparseFloatValueData.data(),
+        HPPL_STREAM_DEFAULT);
   } else if (std::dynamic_pointer_cast<CpuSparseMatrix>(mat)) {
-    std::dynamic_pointer_cast<CpuSparseMatrix>(mat)
-        ->copyFrom(slot.sampleSequenceIdVec.data(),
-                   slot.indices.data(),
-                   slot.sparseFloatValueData.data());
+    std::dynamic_pointer_cast<CpuSparseMatrix>(mat)->copyFrom(
+        slot.sampleSequenceIdVec.data(),
+        slot.indices.data(),
+        slot.sparseFloatValueData.data());
   } else {
     LOG(FATAL) << "Not Supported";
   }

paddle/gserver/dataproviders/PyDataProvider2.cpp

@@ -15,18 +15,18 @@ limitations under the License. */
 #ifndef PADDLE_NO_PYTHON
 
 #include <Python.h>
+#include <numpy/numpyconfig.h>
 #include <stdio.h>
 #include <stdlib.h>
-#include <unordered_set>
 #include <list>
-#include <numpy/numpyconfig.h>
+#include <unordered_set>
 #define NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION
 #include <numpy/ndarrayobject.h>
 
 #include "DataProvider.h"
 
-#include "paddle/utils/PythonUtil.h"
 #include "paddle/utils/Locks.h"
+#include "paddle/utils/PythonUtil.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {
@@ -400,10 +400,9 @@ private:
 
       if (this->loadThread_) {  // wait poolActualSize < poolSize;
         std::unique_lock<std::mutex> l(mtx_);
-        pushCV_.wait(l,
-                     [this, additionalBatchSize] {
-                       return this->poolActualSize_ < poolSize_;
-                     });
+        pushCV_.wait(l, [this, additionalBatchSize] {
+          return this->poolActualSize_ < poolSize_;
+        });
       }
 
       {
@@ -529,12 +528,10 @@ public:
                         // but, loading from cache, cache object should ensure
                         // data pool ready.
       std::unique_lock<std::mutex> l(mtx_);
-      pullCV_.wait(l,
-                   [this, &size] {
-                     return this->poolActualSize_ >=
-                                std::max(size, this->minPoolSize_) ||
-                            callingContexts_.empty();
-                   });
+      pullCV_.wait(l, [this, &size] {
+        return this->poolActualSize_ >= std::max(size, this->minPoolSize_) ||
+               callingContexts_.empty();
+      });
 
       if (unittest::OnPoolFilled) {
         (*unittest::OnPoolFilled)(this->poolActualSize_);

paddle/gserver/evaluators/Evaluator.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
 #include "paddle/gserver/evaluators/Evaluator.h"
+#include "paddle/utils/Stat.h"
 
 #include "paddle/gserver/gradientmachines/NeuralNetwork.h"
 
@@ -842,9 +842,9 @@ void PnpairEvaluator::calc(std::vector<PredictionResult>& predictArray) {
   auto start = predictArray.begin();
   while (start != predictArray.end()) {
     auto end = std::find_if(
-        start + 1,
-        predictArray.end(),
-        [=](const PredictionResult& x) { return x.queryid != start->queryid; });
+        start + 1, predictArray.end(), [=](const PredictionResult& x) {
+          return x.queryid != start->queryid;
+        });
     CHECK(end != start);
     stat(start - predictArray.begin(),
          end - predictArray.begin(),

paddle/gserver/evaluators/Evaluator.h

@@ -14,11 +14,11 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/pserver/ParameterClient2.h"
-#include "paddle/utils/ClassRegistrar.h"
+#include <fstream>
 #include "ModelConfig.pb.h"
 #include "paddle/parameter/Argument.h"
-#include <fstream>
+#include "paddle/pserver/ParameterClient2.h"
+#include "paddle/utils/ClassRegistrar.h"
 
 namespace paddle {
 

paddle/gserver/gradientmachines/GradientMachine.cpp

@@ -14,16 +14,16 @@ limitations under the License. */
 
 #include "GradientMachine.h"
 
-#include "paddle/utils/Logging.h"
 #include <fstream>
+#include "paddle/utils/Logging.h"
 
-#include "hl_gpu.h"
-#include "NeuralNetwork.h"
-#include "ParallelNeuralNetwork.h"
+#include "GradientMachineMode.h"
 #include "MultiGradientMachine.h"
-#include "NeuralNetwork.h"
 #include "MultiNetwork.h"
-#include "GradientMachineMode.h"
+#include "NeuralNetwork.h"
+#include "NeuralNetwork.h"
+#include "ParallelNeuralNetwork.h"
+#include "hl_gpu.h"
 
 namespace paddle {
 

paddle/gserver/gradientmachines/GradientMachine.h

@@ -17,15 +17,15 @@ limitations under the License. */
 #include <iostream>
 #include <vector>
 
-#include "paddle/math/Matrix.h"
-#include "paddle/parameter/Parameter.h"
-#include "paddle/parameter/ParameterUpdaterBase.h"
-#include "paddle/utils/Thread.h"
-#include "TrainerConfig.pb.h"
 #include "ModelConfig.pb.h"
+#include "TrainerConfig.pb.h"
 #include "paddle/gserver/dataproviders/DataProvider.h"
 #include "paddle/gserver/evaluators/Evaluator.h"
 #include "paddle/gserver/layers/Layer.h"
+#include "paddle/math/Matrix.h"
+#include "paddle/parameter/Parameter.h"
+#include "paddle/parameter/ParameterUpdaterBase.h"
+#include "paddle/utils/Thread.h"
 
 namespace paddle {
 /**

paddle/gserver/gradientmachines/MultiGradientMachine.h

@@ -18,9 +18,9 @@ limitations under the License. */
 
 #include "GradientMachine.h"
 
-#include "paddle/utils/Queue.h"
-#include "paddle/utils/Locks.h"
 #include "hl_gpu.h"
+#include "paddle/utils/Locks.h"
+#include "paddle/utils/Queue.h"
 
 namespace paddle {
 

paddle/gserver/gradientmachines/MultiNetwork.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include <algorithm>
 #include "paddle/utils/Stat.h"
 #include "paddle/utils/Util.h"
-#include <algorithm>
 
 #include "MultiNetwork.h"
 

paddle/gserver/gradientmachines/NeuralNetwork.cpp

@@ -14,15 +14,15 @@ limitations under the License. */
 
 #include "paddle/utils/Util.h"
 
-#include "paddle/utils/Logging.h"
 #include "paddle/utils/CustomStackTrace.h"
+#include "paddle/utils/Logging.h"
 
-#include "paddle/utils/Stat.h"
-#include "hl_gpu.h"
+#include "MultiNetwork.h"
 #include "NeuralNetwork.h"
 #include "RecurrentGradientMachine.h"
-#include "MultiNetwork.h"
+#include "hl_gpu.h"
 #include "paddle/gserver/layers/AgentLayer.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 void parameterInitNN(int paramId,

paddle/gserver/gradientmachines/NeuralNetwork.h

@@ -14,18 +14,18 @@ limitations under the License. */
 
 #pragma once
 
-#include <memory>
-#include <map>
 #include <functional>
+#include <map>
+#include <memory>
 
-#include "paddle/utils/ClassRegistrar.h"
-#include "paddle/parameter/Parameter.h"
 #include "ModelConfig.pb.h"
+#include "paddle/gserver/dataproviders/DataProvider.h"
 #include "paddle/gserver/gradientmachines/GradientMachine.h"
 #include "paddle/gserver/layers/CostLayer.h"
 #include "paddle/gserver/layers/DataLayer.h"
-#include "paddle/gserver/dataproviders/DataProvider.h"
 #include "paddle/gserver/layers/Layer.h"
+#include "paddle/parameter/Parameter.h"
+#include "paddle/utils/ClassRegistrar.h"
 
 namespace paddle {
 /*
@@ -57,14 +57,13 @@ void parameterInitNN(int paramId,
 
 class NeuralNetwork : public GradientMachine {
 public:
-  virtual void init(
-      const ModelConfig& config,
-      ParamInitCallback callback = nullptr,
-      const std::vector<ParameterType>&
-          parameterTypes = std::vector<ParameterType>{PARAMETER_VALUE,
-                                                      PARAMETER_GRADIENT,
-                                                      PARAMETER_MOMENTUM},
-      bool useGpu = FLAGS_use_gpu);
+  virtual void init(const ModelConfig& config,
+                    ParamInitCallback callback = nullptr,
+                    const std::vector<ParameterType>& parameterTypes =
+                        std::vector<ParameterType>{PARAMETER_VALUE,
+                                                   PARAMETER_GRADIENT,
+                                                   PARAMETER_MOMENTUM},
+                    bool useGpu = FLAGS_use_gpu);
 
   /**
    * Connect two submodels and

paddle/gserver/gradientmachines/ParallelNeuralNetwork.h

@@ -37,14 +37,13 @@ public:
                         NeuralNetwork *rootNetwork = nullptr)
       : NeuralNetwork(subModelName, rootNetwork) {}
 
-  virtual void init(
-      const ModelConfig &config,
-      ParamInitCallback callback = nullptr,
-      const std::vector<ParameterType>
-          &parameterTypes = std::vector<ParameterType>{PARAMETER_VALUE,
-                                                       PARAMETER_GRADIENT,
-                                                       PARAMETER_MOMENTUM},
-      bool useGpu = FLAGS_use_gpu);
+  virtual void init(const ModelConfig &config,
+                    ParamInitCallback callback = nullptr,
+                    const std::vector<ParameterType> &parameterTypes =
+                        std::vector<ParameterType>{PARAMETER_VALUE,
+                                                   PARAMETER_GRADIENT,
+                                                   PARAMETER_MOMENTUM},
+                    bool useGpu = FLAGS_use_gpu);
 
   virtual void forward(const std::vector<Argument> &inArgs,
                        std::vector<Argument> *outArgs,

paddle/gserver/gradientmachines/RecurrentGradientMachine.cpp

@@ -12,17 +12,17 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
-#include "paddle/utils/Util.h"
-#include "paddle/utils/Flags.h"
+#include "RecurrentGradientMachine.h"
+#include <dlfcn.h>
 #include <algorithm>
+#include <cmath>
 #include <functional>
-#include <dlfcn.h>
 #include <limits>
-#include <cmath>
-#include "RecurrentGradientMachine.h"
 #include "NeuralNetwork.h"
 #include "paddle/gserver/layers/AgentLayer.h"
+#include "paddle/utils/Flags.h"
+#include "paddle/utils/Stat.h"
+#include "paddle/utils/Util.h"
 
 P_DEFINE_string(diy_beam_search_prob_so, "", "the diy beam search cost so");
 
@@ -78,20 +78,22 @@ static inline SymbolType loadDiySymbol(const char* symbolName) {
   return reinterpret_cast<SymbolType>(sym);
 }
 
-static InitFunction __init__diy_prob_method([] {
-  std::string soName = FLAGS_diy_beam_search_prob_so;
-  if (!soName.empty()) {
-    gDiyProbHandle = dlopen(soName.c_str(), RTLD_LAZY);
-    CHECK(gDiyProbHandle) << "Cannot Open DIY Prob So " << soName;
-    atexit(exit_diy_prob);
-    gDiyProbMethod =
-        loadDiySymbol<decltype(gDiyProbMethod)>(DIY_CALC_PROB_SYMBOL_NAME);
-    gDiyProbStart =
-        loadDiySymbol<decltype(gDiyProbStart)>(DIY_START_CALC_PROB_SYMBOL_NAME);
-    gDiyProbStop =
-        loadDiySymbol<decltype(gDiyProbStop)>(DIY_FINISH_CALC_PROB_SYMBOL_NAME);
-  }
-}, std::numeric_limits<int>::max());
+static InitFunction __init__diy_prob_method(
+    [] {
+      std::string soName = FLAGS_diy_beam_search_prob_so;
+      if (!soName.empty()) {
+        gDiyProbHandle = dlopen(soName.c_str(), RTLD_LAZY);
+        CHECK(gDiyProbHandle) << "Cannot Open DIY Prob So " << soName;
+        atexit(exit_diy_prob);
+        gDiyProbMethod =
+            loadDiySymbol<decltype(gDiyProbMethod)>(DIY_CALC_PROB_SYMBOL_NAME);
+        gDiyProbStart = loadDiySymbol<decltype(gDiyProbStart)>(
+            DIY_START_CALC_PROB_SYMBOL_NAME);
+        gDiyProbStop = loadDiySymbol<decltype(gDiyProbStop)>(
+            DIY_FINISH_CALC_PROB_SYMBOL_NAME);
+      }
+    },
+    std::numeric_limits<int>::max());
 
 class BeamSearchControlCallbacks {
 public:
@@ -1281,10 +1283,9 @@ void RecurrentGradientMachine::beamSearch(size_t batchSize) {
       std::vector<std::vector<int>*> prefixes;
       prefixes.resize(paths.size());
       std::transform(
-          paths.begin(),
-          paths.end(),
-          prefixes.begin(),
-          [](const Path& p) { return const_cast<std::vector<int>*>(&p.ids); });
+          paths.begin(), paths.end(), prefixes.begin(), [](const Path& p) {
+            return const_cast<std::vector<int>*>(&p.ids);
+          });
       beamSearchCtrlCallbacks_->beamSearchCandidateAdjust(
           prefixes, frames_[machineCur].get(), i);
     }

paddle/gserver/gradientmachines/RecurrentGradientMachine.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
+#include <functional>
 #include "GradientMachine.h"
 #include "NeuralNetwork.h"
-#include <functional>
 
 #include "paddle/utils/Locks.h"
 

paddle/gserver/layers/BatchNormBaseLayer.cpp

@@ -12,10 +12,10 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
-#include "Layer.h"
 #include "BatchNormBaseLayer.h"
 #include "BatchNormalizationLayer.h"
+#include "Layer.h"
+#include "paddle/utils/Stat.h"
 #ifndef PADDLE_ONLY_CPU
 #include "CudnnBatchNormLayer.h"
 #endif

paddle/gserver/layers/BatchNormBaseLayer.h

@@ -14,8 +14,8 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/utils/Stat.h"
 #include "Layer.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/BatchNormalizationLayer.h

@@ -14,8 +14,8 @@ limitations under the License. */
 
 #pragma once
 
-#include "Layer.h"
 #include "BatchNormBaseLayer.h"
+#include "Layer.h"
 
 namespace paddle {
 

paddle/gserver/layers/ConcatenateLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
 #include "Layer.h"
 #include "Projection.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/ContextProjection.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
 #include "ContextProjection.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/ConvBaseLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Logging.h"
 #include "ConvBaseLayer.h"
 #include "paddle/math/MathUtils.h"
+#include "paddle/utils/Logging.h"
 namespace paddle {
 
 bool ConvBaseLayer::init(const LayerMap& layerMap,

paddle/gserver/layers/ConvOperator.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/math/Matrix.h"
-#include "paddle/math/MathUtils.h"
 #include "Operator.h"
+#include "paddle/math/MathUtils.h"
+#include "paddle/math/Matrix.h"
 
 namespace paddle {
 

paddle/gserver/layers/ConvProjection.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
 #include "ConvProjection.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/ConvShiftLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Logging.h"
 #include "Layer.h"
 #include "paddle/math/Matrix.h"
+#include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {

paddle/gserver/layers/ConvexCombinationLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Logging.h"
 #include "Layer.h"
 #include "paddle/math/Matrix.h"
+#include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {

paddle/gserver/layers/CosSimVecMatLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Logging.h"
 #include "Layer.h"
 #include "paddle/math/Matrix.h"
+#include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {

paddle/gserver/layers/CostLayer.cpp

@@ -12,11 +12,11 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <memory>
+#include "CostLayer.h"
 #include <algorithm>
-#include "paddle/utils/Logging.h"
 #include <cmath>
-#include "CostLayer.h"
+#include <memory>
+#include "paddle/utils/Logging.h"
 
 #include "paddle/math/SparseMatrix.h"
 

paddle/gserver/layers/CudnnBatchNormLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
-#include "Layer.h"
 #include "CudnnBatchNormLayer.h"
+#include "Layer.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/CudnnBatchNormLayer.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/utils/Stat.h"
-#include "Layer.h"
 #include "BatchNormBaseLayer.h"
+#include "Layer.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/CudnnConvLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "CudnnConvLayer.h"
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
-#include "CudnnConvLayer.h"
 
 namespace paddle {
 

paddle/gserver/layers/CudnnConvLayer.h

@@ -14,10 +14,10 @@ limitations under the License. */
 
 #pragma once
 
+#include <vector>
 #include "ConvBaseLayer.h"
-#include "paddle/math/Matrix.h"
 #include "Projection.h"
-#include <vector>
+#include "paddle/math/Matrix.h"
 
 namespace paddle {
 

paddle/gserver/layers/CudnnPoolLayer.cpp

@@ -12,10 +12,10 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "CudnnPoolLayer.h"
+#include "paddle/math/Matrix.h"
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
-#include "paddle/math/Matrix.h"
-#include "CudnnPoolLayer.h"
 
 namespace paddle {
 

paddle/gserver/layers/EosIdCheckLayer.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Logging.h"
 #include "Layer.h"
+#include "paddle/utils/Logging.h"
 
 namespace paddle {
 /**

paddle/gserver/layers/ExpandConvBaseLayer.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
+#include <vector>
 #include "ConvBaseLayer.h"
 #include "paddle/math/Matrix.h"
-#include <vector>
 
 namespace paddle {
 

paddle/gserver/layers/ExpandConvLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "ExpandConvLayer.h"
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
-#include "ExpandConvLayer.h"
 
 namespace paddle {
 

paddle/gserver/layers/ExpandConvLayer.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/math/Matrix.h"
 #include <vector>
 #include "ExpandConvBaseLayer.h"
+#include "paddle/math/Matrix.h"
 
 namespace paddle {
 

paddle/gserver/layers/ExpandConvTransLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include "ExpandConvTransLayer.h"
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
-#include "ExpandConvTransLayer.h"
 
 /* The implementation of the convTransLayer is basically a swap of forward and
  * backward of the original convLayer.

paddle/gserver/layers/ExpandConvTransLayer.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/math/Matrix.h"
 #include <vector>
 #include "ExpandConvBaseLayer.h"
+#include "paddle/math/Matrix.h"
 
 namespace paddle {
 

paddle/gserver/layers/FullyConnectedLayer.cpp

@@ -13,11 +13,11 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "FullyConnectedLayer.h"
+#include <algorithm>
+#include <vector>
+#include "paddle/math/SparseMatrix.h"
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
-#include "paddle/math/SparseMatrix.h"
-#include <vector>
-#include <algorithm>
 
 namespace paddle {
 

paddle/gserver/layers/GatedRecurrentLayer.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "Layer.h"
 #include "GatedRecurrentLayer.h"
+#include "Layer.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {
@@ -386,8 +386,9 @@ void GatedRecurrentLayer::backwardBatch(int batchSize, MatrixPtr inputGrad) {
       {
         batchSize = outputGradTmp->getHeight();
         gruValue.prevOutValue =
-            (n == 0 ? nullptr : (batchValue_->getBatchValue(n - 1, batchSize))
-                                    ->getData());
+            (n == 0
+                 ? nullptr
+                 : (batchValue_->getBatchValue(n - 1, batchSize))->getData());
         gruGrad.prevOutGrad =
             (n == 0 ? nullptr
                     : (batchGrad_->getBatchValue(n - 1, batchSize))->getData());

paddle/gserver/layers/GatedRecurrentLayer.h

@@ -14,10 +14,10 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/math/Matrix.h"
-#include "SequenceToBatch.h"
 #include "GruCompute.h"
 #include "Layer.h"
+#include "SequenceToBatch.h"
+#include "paddle/math/Matrix.h"
 
 namespace paddle {
 

paddle/gserver/layers/GruCompute.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Util.h"
 #include "GruCompute.h"
 #include "hl_recurrent_apply.cuh"
+#include "paddle/utils/Util.h"
 
 namespace paddle {
 

paddle/gserver/layers/GruCompute.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/utils/TypeDefs.h"
 #include "ModelConfig.pb.h"
 #include "hl_gpu.h"
+#include "paddle/utils/TypeDefs.h"
 
 namespace paddle {
 

paddle/gserver/layers/GruStepLayer.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "Layer.h"
 #include "GruCompute.h"
+#include "Layer.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {

paddle/gserver/layers/IdentityProjection.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Stat.h"
 #include "Projection.h"
+#include "paddle/utils/Stat.h"
 
 namespace paddle {
 

paddle/gserver/layers/InterpolationLayer.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Logging.h"
 #include "Layer.h"
 #include "paddle/math/Matrix.h"
+#include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
 
 namespace paddle {

paddle/gserver/layers/Layer.cpp

@@ -14,15 +14,15 @@ limitations under the License. */
 
 #include "paddle/utils/Util.h"
 
-#include "paddle/utils/Logging.h"
 #include "paddle/math/SparseMatrix.h"
+#include "paddle/utils/Logging.h"
 
 #include "AddtoLayer.h"
+#include "CRFLayer.h"
 #include "CosSimLayer.h"
 #include "CostLayer.h"
-#include "ExpandConvLayer.h"
-#include "CRFLayer.h"
 #include "DataLayer.h"
+#include "ExpandConvLayer.h"
 #include "FullyConnectedLayer.h"
 #include "HierarchicalSigmoidLayer.h"
 #include "MaxLayer.h"

paddle/gserver/layers/Layer.h

@@ -14,18 +14,18 @@ limitations under the License. */
 
 #pragma once
 
-#include <memory>
-#include <functional>
 #include <paddle/parameter/Argument.h>
-#include "paddle/utils/ClassRegistrar.h"
+#include <functional>
+#include <memory>
+#include "ModelConfig.pb.h"
 #include "paddle/math/CpuSparseMatrix.h"
 #include "paddle/parameter/Parameter.h"
+#include "paddle/utils/ClassRegistrar.h"
 #include "paddle/utils/Util.h"
-#include "ModelConfig.pb.h"
 
-#include "paddle/gserver/activations/ActivationFunction.h"
 #include <paddle/parameter/ParallelParameter.h>
 #include <paddle/parameter/Weight.h>
+#include "paddle/gserver/activations/ActivationFunction.h"
 
 /// Macro for registering a layer type.
 /// Example: REGISTER_LAYER(crf_error, CRFDecodingErrorLayer);

paddle/gserver/layers/LinearChainCRF.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <algorithm>
 #include "LinearChainCRF.h"
+#include <algorithm>
 
 namespace paddle {
 

paddle/gserver/layers/LinearChainCTC.cpp

@@ -12,8 +12,8 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <math.h>
 #include "LinearChainCTC.h"
+#include <math.h>
 #include <limits>
 
 namespace paddle {

paddle/gserver/layers/LstmCompute.cpp

@@ -12,9 +12,9 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/utils/Util.h"
-#include "hl_recurrent_apply.cuh"
 #include "LstmCompute.h"
+#include "hl_recurrent_apply.cuh"
+#include "paddle/utils/Util.h"
 
 namespace paddle {
 

paddle/gserver/layers/LstmCompute.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/utils/TypeDefs.h"
 #include "ModelConfig.pb.h"
 #include "hl_gpu.h"
+#include "paddle/utils/TypeDefs.h"
 
 namespace paddle {