Merge branch 'develop' of github.com:baidu/Paddle into feature/serialize_deserialize_in_parameters

.travis.yml

@@ -4,22 +4,14 @@ cache:
     - $HOME/third_party
     - $HOME/.ccache
     - $HOME/.cache/pip
-    - $HOME/Library/Caches/Homebrew
 sudo: required
 dist: trusty
 os:
   - linux
-  - osx
 env:
   - JOB=DOCS
   - JOB=BUILD_AND_TEST
   - JOB=PRE_COMMIT
-matrix:
-  exclude:
-    - os: osx
-      env: JOB=DOCS  # Only generate documentation in linux.
-    - os: osx
-      env: JOB=PRE_COMMIT # Only check pre-commit hook in linux
 
 addons:
   apt:
@@ -53,7 +45,6 @@ before_install:
         fi
       fi
     fi
-  - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then paddle/scripts/travis/before_install.osx.sh; fi
   - if [[ "$JOB" == "PRE_COMMIT" ]]; then sudo ln -s /usr/bin/clang-format-3.8 /usr/bin/clang-format; fi
   # Paddle is using protobuf 3.1 currently. Protobuf 3.2 breaks the compatibility. So we specify the python 
   # protobuf version.

cmake/FindSphinx.cmake

@@ -72,7 +72,7 @@ function( Sphinx_add_target target_name builder conf cache source destination )
     ${source}
     ${destination}
     COMMENT "Generating sphinx documentation: ${builder}"
-    COMMAND ln -sf ${destination}/index_*.html ${destination}/index.html
+    COMMAND cd ${destination} && ln -s ./index_*.html index.html
     )
 
   set_property(

cmake/coverallsGcovJsons.cmake

@@ -110,14 +110,13 @@ endmacro()
 
 # Get the coverage data.
 file(GLOB_RECURSE GCDA_FILES "${COV_PATH}" "*.gcda")
-message("GCDA files:")
+message("Process GCDA files:")
+message("===============================")
 
 # Get a list of all the object directories needed by gcov
 # (The directories the .gcda files and .o files are found in)
 # and run gcov on those.
 foreach(GCDA ${GCDA_FILES})
-	message("Process: ${GCDA}")
-	message("------------------------------------------------------------------------------")
 	get_filename_component(GCDA_DIR ${GCDA} PATH)
 
 	#
@@ -135,7 +134,7 @@ foreach(GCDA ${GCDA_FILES})
 	# If -p is not specified then the file is named only "the_file.c.gcov"
 	#
 	execute_process(
-		COMMAND ${GCOV_EXECUTABLE} -p -o ${GCDA_DIR} ${GCDA}
+		COMMAND "${GCOV_EXECUTABLE} -p -o ${GCDA_DIR} ${GCDA} >/dev/null"
 		WORKING_DIRECTORY ${GCDA_DIR}
 	)
 endforeach()
@@ -383,7 +382,6 @@ foreach(NOT_COVERED_SRC ${COVERAGE_SRCS_REMAINING})
 	set(GCOV_FILE_COVERAGE "${GCOV_FILE_COVERAGE}]")
 
 	# Generate the final JSON for this file.
-	message("Generate JSON for non-gcov file: ${NOT_COVERED_SRC}...")
 	string(CONFIGURE ${SRC_FILE_TEMPLATE} FILE_JSON)
 	set(JSON_GCOV_FILES "${JSON_GCOV_FILES}${FILE_JSON}, ")
 endforeach()

cmake/external/protobuf.cmake

@@ -14,46 +14,50 @@
 
 INCLUDE(ExternalProject)
 
-SET(PROTOBUF_SOURCES_DIR ${THIRD_PARTY_PATH}/protobuf)
-SET(PROTOBUF_INSTALL_DIR ${THIRD_PARTY_PATH}/install/protobuf)
-SET(PROTOBUF_INCLUDE_DIR "${PROTOBUF_INSTALL_DIR}/include" CACHE PATH "protobuf include directory." FORCE)
+FIND_PACKAGE(Protobuf)
 
-INCLUDE_DIRECTORIES(${PROTOBUF_INCLUDE_DIR})
+IF(NOT PROTOBUF_FOUND)
+    SET(PROTOBUF_SOURCES_DIR ${THIRD_PARTY_PATH}/protobuf)
+    SET(PROTOBUF_INSTALL_DIR ${THIRD_PARTY_PATH}/install/protobuf)
+    SET(PROTOBUF_INCLUDE_DIR "${PROTOBUF_INSTALL_DIR}/include" CACHE PATH "protobuf include directory." FORCE)
+
+    IF(WIN32)
+        SET(PROTOBUF_LITE_LIBRARY
+            "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf-lite.lib" CACHE FILEPATH "protobuf lite library." FORCE)
+        SET(PROTOBUF_LIBRARY
+            "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf.lib" CACHE FILEPATH "protobuf library." FORCE)
+        SET(PROTOBUF_PROTOC_LIBRARY
+            "${PROTOBUF_INSTALL_DIR}/lib/libprotoc.lib" CACHE FILEPATH "protoc library." FORCE)
+        SET(PROTOBUF_PROTOC_EXECUTABLE "${PROTOBUF_INSTALL_DIR}/bin/protoc.exe" CACHE FILEPATH "protobuf executable." FORCE)
+    ELSE(WIN32)
+        SET(PROTOBUF_LITE_LIBRARY
+            "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf-lite.a" CACHE FILEPATH "protobuf lite library." FORCE)
+        SET(PROTOBUF_LIBRARY
+            "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf.a" CACHE FILEPATH "protobuf library." FORCE)
+        SET(PROTOBUF_PROTOC_LIBRARY
+            "${PROTOBUF_INSTALL_DIR}/lib/libprotoc.a" CACHE FILEPATH "protoc library." FORCE)
+        SET(PROTOBUF_PROTOC_EXECUTABLE "${PROTOBUF_INSTALL_DIR}/bin/protoc" CACHE FILEPATH "protobuf executable." FORCE)
+    ENDIF(WIN32)
 
-IF(WIN32)
-  SET(PROTOBUF_LITE_LIBRARY
-        "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf-lite.lib" CACHE FILEPATH "protobuf lite library." FORCE)
-  SET(PROTOBUF_LIBRARY
-        "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf.lib" CACHE FILEPATH "protobuf library." FORCE)
-  SET(PROTOBUF_PROTOC_LIBRARY
-        "${PROTOBUF_INSTALL_DIR}/lib/libprotoc.lib" CACHE FILEPATH "protoc library." FORCE)
-  SET(PROTOBUF_PROTOC_EXECUTABLE "${PROTOBUF_INSTALL_DIR}/bin/protoc.exe" CACHE FILEPATH "protobuf executable." FORCE)
-ELSE(WIN32)
-  SET(PROTOBUF_LITE_LIBRARY
-        "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf-lite.a" CACHE FILEPATH "protobuf lite library." FORCE)
-  SET(PROTOBUF_LIBRARY
-        "${PROTOBUF_INSTALL_DIR}/lib/libprotobuf.a" CACHE FILEPATH "protobuf library." FORCE)
-  SET(PROTOBUF_PROTOC_LIBRARY
-        "${PROTOBUF_INSTALL_DIR}/lib/libprotoc.a" CACHE FILEPATH "protoc library." FORCE)
-  SET(PROTOBUF_PROTOC_EXECUTABLE "${PROTOBUF_INSTALL_DIR}/bin/protoc" CACHE FILEPATH "protobuf executable." FORCE)
-ENDIF(WIN32)
+    ExternalProject_Add(
+        protobuf
+        ${EXTERNAL_PROJECT_LOG_ARGS}
+        PREFIX          ${PROTOBUF_SOURCES_DIR}
+        UPDATE_COMMAND  ""
+        DEPENDS         zlib
+        GIT_REPOSITORY  "https://github.com/google/protobuf.git"
+        GIT_TAG         "9f75c5aa851cd877fb0d93ccc31b8567a6706546"
+        CONFIGURE_COMMAND
+        ${CMAKE_COMMAND} ${PROTOBUF_SOURCES_DIR}/src/protobuf/cmake
+        -Dprotobuf_BUILD_TESTS=OFF
+        -DZLIB_ROOT:FILEPATH=${ZLIB_ROOT}
+        -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+        -DCMAKE_BUILD_TYPE=Release
+        -DCMAKE_INSTALL_PREFIX=${PROTOBUF_INSTALL_DIR}
+        -DCMAKE_INSTALL_LIBDIR=lib
+    )
 
-ExternalProject_Add(
-  protobuf
-  ${EXTERNAL_PROJECT_LOG_ARGS}
-  PREFIX          ${PROTOBUF_SOURCES_DIR}
-  UPDATE_COMMAND  ""
-  DEPENDS         zlib
-  GIT_REPOSITORY  "https://github.com/google/protobuf.git"
-  GIT_TAG         "9f75c5aa851cd877fb0d93ccc31b8567a6706546"
-  CONFIGURE_COMMAND
-    ${CMAKE_COMMAND} ${PROTOBUF_SOURCES_DIR}/src/protobuf/cmake
-    -Dprotobuf_BUILD_TESTS=OFF
-    -DZLIB_ROOT:FILEPATH=${ZLIB_ROOT}
-    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    -DCMAKE_BUILD_TYPE=Release
-    -DCMAKE_INSTALL_PREFIX=${PROTOBUF_INSTALL_DIR}
-    -DCMAKE_INSTALL_LIBDIR=lib
-)
+    LIST(APPEND external_project_dependencies protobuf)
+ENDIF(NOT PROTOBUF_FOUND)
 
-LIST(APPEND external_project_dependencies protobuf)
+INCLUDE_DIRECTORIES(${PROTOBUF_INCLUDE_DIR})

cmake/external/python.cmake

@@ -221,7 +221,3 @@ ENDIF(PYTHONLIBS_FOUND AND PYTHONINTERP_FOUND)
 
 INCLUDE_DIRECTORIES(${PYTHON_INCLUDE_DIR})
 INCLUDE_DIRECTORIES(${PYTHON_NUMPY_INCLUDE_DIR})
-
-MESSAGE("[Paddle] Python Executable: ${PYTHON_EXECUTABLE}")
-MESSAGE("[Paddle] Python Include: ${PYTHON_INCLUDE_DIRS}")
-MESSAGE("[Paddle] Python Libraries: ${PYTHON_LIBRARIES}")

demo/image_classification/api_v2_resnet.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+import paddle.v2 as paddle
+
+__all__ = ['resnet_cifar10']
+
+
+def conv_bn_layer(input,
+                  ch_out,
+                  filter_size,
+                  stride,
+                  padding,
+                  active_type=paddle.activation.Relu(),
+                  ch_in=None):
+    tmp = paddle.layer.img_conv(
+        input=input,
+        filter_size=filter_size,
+        num_channels=ch_in,
+        num_filters=ch_out,
+        stride=stride,
+        padding=padding,
+        act=paddle.activation.Linear(),
+        bias_attr=False)
+    return paddle.layer.batch_norm(input=tmp, act=active_type)
+
+
+def shortcut(ipt, n_in, n_out, stride):
+    if n_in != n_out:
+        return conv_bn_layer(ipt, n_out, 1, stride, 0,
+                             paddle.activation.Linear())
+    else:
+        return ipt
+
+
+def basicblock(ipt, ch_out, stride):
+    ch_in = ch_out * 2
+    tmp = conv_bn_layer(ipt, ch_out, 3, stride, 1)
+    tmp = conv_bn_layer(tmp, ch_out, 3, 1, 1, paddle.activation.Linear())
+    short = shortcut(ipt, ch_in, ch_out, stride)
+    return paddle.layer.addto(input=[tmp, short], act=paddle.activation.Relu())
+
+
+def layer_warp(block_func, ipt, features, count, stride):
+    tmp = block_func(ipt, features, stride)
+    for i in range(1, count):
+        tmp = block_func(tmp, features, 1)
+    return tmp
+
+
+def resnet_cifar10(ipt, depth=32):
+    # depth should be one of 20, 32, 44, 56, 110, 1202
+    assert (depth - 2) % 6 == 0
+    n = (depth - 2) / 6
+    nStages = {16, 64, 128}
+    conv1 = conv_bn_layer(
+        ipt, ch_in=3, ch_out=16, filter_size=3, stride=1, padding=1)
+    res1 = layer_warp(basicblock, conv1, 16, n, 1)
+    res2 = layer_warp(basicblock, res1, 32, n, 2)
+    res3 = layer_warp(basicblock, res2, 64, n, 2)
+    pool = paddle.layer.img_pool(
+        input=res3, pool_size=8, stride=1, pool_type=paddle.pooling.Avg())
+    return pool

demo/image_classification/api_v2_train.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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
+
+import sys
+import paddle.v2 as paddle
+from api_v2_vgg import vgg_bn_drop
+from api_v2_resnet import resnet_cifar10
+
+
+def main():
+    datadim = 3 * 32 * 32
+    classdim = 10
+
+    # PaddlePaddle init
+    paddle.init(use_gpu=True, trainer_count=1)
+
+    image = paddle.layer.data(
+        name="image", type=paddle.data_type.dense_vector(datadim))
+
+    # Add neural network config
+    # option 1. resnet
+    net = resnet_cifar10(image, depth=32)
+    # option 2. vgg
+    # net = vgg_bn_drop(image)
+
+    out = paddle.layer.fc(input=net,
+                          size=classdim,
+                          act=paddle.activation.Softmax())
+
+    lbl = paddle.layer.data(
+        name="label", type=paddle.data_type.integer_value(classdim))
+    cost = paddle.layer.classification_cost(input=out, label=lbl)
+
+    # Create parameters
+    parameters = paddle.parameters.create(cost)
+
+    # Create optimizer
+    momentum_optimizer = paddle.optimizer.Momentum(
+        momentum=0.9,
+        regularization=paddle.optimizer.L2Regularization(rate=0.0002 * 128),
+        learning_rate=0.1 / 128.0,
+        learning_rate_decay_a=0.1,
+        learning_rate_decay_b=50000 * 100,
+        learning_rate_schedule='discexp',
+        batch_size=128)
+
+    # End batch and end pass event handler
+    def event_handler(event):
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 100 == 0:
+                print "\nPass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics)
+            else:
+                sys.stdout.write('.')
+                sys.stdout.flush()
+        if isinstance(event, paddle.event.EndPass):
+            result = trainer.test(
+                reader=paddle.reader.batched(
+                    paddle.dataset.cifar.test10(), batch_size=128),
+                reader_dict={'image': 0,
+                             'label': 1})
+            print "\nTest with Pass %d, %s" % (event.pass_id, result.metrics)
+
+    # Create trainer
+    trainer = paddle.trainer.SGD(cost=cost,
+                                 parameters=parameters,
+                                 update_equation=momentum_optimizer)
+    trainer.train(
+        reader=paddle.reader.batched(
+            paddle.reader.shuffle(
+                paddle.dataset.cifar.train10(), buf_size=50000),
+            batch_size=128),
+        num_passes=5,
+        event_handler=event_handler,
+        reader_dict={'image': 0,
+                     'label': 1})
+
+
+if __name__ == '__main__':
+    main()

demo/image_classification/api_v2_vgg.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+import paddle.v2 as paddle
+
+__all__ = ['vgg_bn_drop']
+
+
+def vgg_bn_drop(input):
+    def conv_block(ipt, num_filter, groups, dropouts, num_channels=None):
+        return paddle.networks.img_conv_group(
+            input=ipt,
+            num_channels=num_channels,
+            pool_size=2,
+            pool_stride=2,
+            conv_num_filter=[num_filter] * groups,
+            conv_filter_size=3,
+            conv_act=paddle.activation.Relu(),
+            conv_with_batchnorm=True,
+            conv_batchnorm_drop_rate=dropouts,
+            pool_type=paddle.pooling.Max())
+
+    conv1 = conv_block(input, 64, 2, [0.3, 0], 3)
+    conv2 = conv_block(conv1, 128, 2, [0.4, 0])
+    conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0])
+    conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0])
+    conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0])
+
+    drop = paddle.layer.dropout(input=conv5, dropout_rate=0.5)
+    fc1 = paddle.layer.fc(input=drop, size=512, act=paddle.activation.Linear())
+    bn = paddle.layer.batch_norm(
+        input=fc1,
+        act=paddle.activation.Relu(),
+        layer_attr=paddle.attr.Extra(drop_rate=0.5))
+    fc2 = paddle.layer.fc(input=bn, size=512, act=paddle.activation.Linear())
+    return fc2

demo/introduction/api_train_v2.py

+import paddle.v2 as paddle
+import paddle.v2.dataset.uci_housing as uci_housing
+
+
+def main():
+    # init
+    paddle.init(use_gpu=False, trainer_count=1)
+
+    # network config
+    x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))
+    y_predict = paddle.layer.fc(input=x,
+                                param_attr=paddle.attr.Param(name='w'),
+                                size=1,
+                                act=paddle.activation.Linear(),
+                                bias_attr=paddle.attr.Param(name='b'))
+    y = paddle.layer.data(name='y', type=paddle.data_type.dense_vector(1))
+    cost = paddle.layer.regression_cost(input=y_predict, label=y)
+
+    # create parameters
+    parameters = paddle.parameters.create(cost)
+
+    # create optimizer
+    optimizer = paddle.optimizer.Momentum(momentum=0)
+
+    trainer = paddle.trainer.SGD(cost=cost,
+                                 parameters=parameters,
+                                 update_equation=optimizer)
+
+    # event_handler to print training and testing info
+    def event_handler(event):
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 100 == 0:
+                print "Pass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics)
+
+        if isinstance(event, paddle.event.EndPass):
+            result = trainer.test(
+                reader=paddle.reader.batched(
+                    uci_housing.test(), batch_size=2),
+                reader_dict={'x': 0,
+                             'y': 1})
+            if event.pass_id % 10 == 0:
+                print "Test %d, %s" % (event.pass_id, result.metrics)
+
+    # training
+    trainer.train(
+        reader=paddle.reader.batched(
+            paddle.reader.shuffle(
+                uci_housing.train(), buf_size=500),
+            batch_size=2),
+        reader_dict={'x': 0,
+                     'y': 1},
+        event_handler=event_handler,
+        num_passes=30)
+
+
+if __name__ == '__main__':
+    main()

demo/mnist/api_train_v2.py

@@ -2,6 +2,59 @@ import paddle.v2 as paddle
 import cPickle
 
 
+def softmax_regression(img):
+    predict = paddle.layer.fc(input=img,
+                              size=10,
+                              act=paddle.activation.Softmax())
+    return predict
+
+
+def multilayer_perceptron(img):
+    # The first fully-connected layer
+    hidden1 = paddle.layer.fc(input=img, size=128, act=paddle.activation.Relu())
+    # The second fully-connected layer and the according activation function
+    hidden2 = paddle.layer.fc(input=hidden1,
+                              size=64,
+                              act=paddle.activation.Relu())
+    # The thrid fully-connected layer, note that the hidden size should be 10,
+    # which is the number of unique digits
+    predict = paddle.layer.fc(input=hidden2,
+                              size=10,
+                              act=paddle.activation.Softmax())
+    return predict
+
+
+def convolutional_neural_network(img):
+    # first conv layer
+    conv_pool_1 = paddle.networks.simple_img_conv_pool(
+        input=img,
+        filter_size=5,
+        num_filters=20,
+        num_channel=1,
+        pool_size=2,
+        pool_stride=2,
+        act=paddle.activation.Tanh())
+    # second conv layer
+    conv_pool_2 = paddle.networks.simple_img_conv_pool(
+        input=conv_pool_1,
+        filter_size=5,
+        num_filters=50,
+        num_channel=20,
+        pool_size=2,
+        pool_stride=2,
+        act=paddle.activation.Tanh())
+    # The first fully-connected layer
+    fc1 = paddle.layer.fc(input=conv_pool_2,
+                          size=128,
+                          act=paddle.activation.Tanh())
+    # The softmax layer, note that the hidden size should be 10,
+    # which is the number of unique digits
+    predict = paddle.layer.fc(input=fc1,
+                              size=10,
+                              act=paddle.activation.Softmax())
+    return predict
+
+
 def main():
     paddle.init(use_gpu=False, trainer_count=1)
 
@@ -10,12 +63,14 @@ def main():
         name='pixel', type=paddle.data_type.dense_vector(784))
     label = paddle.layer.data(
         name='label', type=paddle.data_type.integer_value(10))
-    hidden1 = paddle.layer.fc(input=images, size=200)
-    hidden2 = paddle.layer.fc(input=hidden1, size=200)
-    inference = paddle.layer.fc(input=hidden2,
-                                size=10,
-                                act=paddle.activation.Softmax())
-    cost = paddle.layer.classification_cost(input=inference, label=label)
+
+    # Here we can build the prediction network in different ways. Please
+    # choose one by uncomment corresponding line.
+    predict = softmax_regression(images)
+    #predict = multilayer_perceptron(images)
+    #predict = convolutional_neural_network(images)
+
+    cost = paddle.layer.classification_cost(input=predict, label=label)
 
     try:
         with open('params.pkl', 'r') as f:
@@ -23,11 +78,16 @@ def main():
     except IOError:
         parameters = paddle.parameters.create(cost)
 
-    adam_optimizer = paddle.optimizer.Adam(learning_rate=0.01)
+    optimizer = paddle.optimizer.Momentum(
+        learning_rate=0.1 / 128.0,
+        momentum=0.9,
+        regularization=paddle.optimizer.L2Regularization(rate=0.0005 * 128))
 
     trainer = paddle.trainer.SGD(cost=cost,
                                  parameters=parameters,
-                                 update_equation=adam_optimizer)
+                                 update_equation=optimizer)
+
+    lists = []
 
     def event_handler(event):
         if isinstance(event, paddle.event.EndIteration):
@@ -43,20 +103,31 @@ def main():
                     cPickle.dump(
                         parameters, f, protocol=cPickle.HIGHEST_PROTOCOL)
 
-        else:
-            pass
+        elif isinstance(event, paddle.event.EndPass):
+            result = trainer.test(reader=paddle.reader.batched(
+                paddle.dataset.mnist.test(), batch_size=128))
+            print "Test with Pass %d, Cost %f, %s\n" % (
+                event.pass_id, result.cost, result.metrics)
+            lists.append((event.pass_id, result.cost,
+                          result.metrics['classification_error_evaluator']))
 
     trainer.train(
         reader=paddle.reader.batched(
             paddle.reader.shuffle(
                 paddle.dataset.mnist.train(), buf_size=8192),
-            batch_size=32),
-        event_handler=event_handler)
+            batch_size=128),
+        event_handler=event_handler,
+        num_passes=100)
+
+    # find the best pass
+    best = sorted(lists, key=lambda list: float(list[1]))[0]
+    print 'Best pass is %s, testing Avgcost is %s' % (best[0], best[1])
+    print 'The classification accuracy is %.2f%%' % (100 - float(best[2]) * 100)
 
     # output is a softmax layer. It returns probabilities.
     # Shape should be (100, 10)
     probs = paddle.infer(
-        output=inference,
+        output=predict,
         parameters=parameters,
         reader=paddle.reader.batched(
             paddle.reader.firstn(

demo/semantic_role_labeling/api_train_v2.py

+import sys
+import math
+import numpy as np
+import paddle.v2 as paddle
+import paddle.v2.dataset.conll05 as conll05
+
+
+def db_lstm():
+    word_dict, verb_dict, label_dict = conll05.get_dict()
+    word_dict_len = len(word_dict)
+    label_dict_len = len(label_dict)
+    pred_len = len(verb_dict)
+
+    mark_dict_len = 2
+    word_dim = 32
+    mark_dim = 5
+    hidden_dim = 512
+    depth = 8
+
+    #8 features
+    def d_type(size):
+        return paddle.data_type.integer_value_sequence(size)
+
+    word = paddle.layer.data(name='word_data', type=d_type(word_dict_len))
+    predicate = paddle.layer.data(name='verb_data', type=d_type(pred_len))
+
+    ctx_n2 = paddle.layer.data(name='ctx_n2_data', type=d_type(word_dict_len))
+    ctx_n1 = paddle.layer.data(name='ctx_n1_data', type=d_type(word_dict_len))
+    ctx_0 = paddle.layer.data(name='ctx_0_data', type=d_type(word_dict_len))
+    ctx_p1 = paddle.layer.data(name='ctx_p1_data', type=d_type(word_dict_len))
+    ctx_p2 = paddle.layer.data(name='ctx_p2_data', type=d_type(word_dict_len))
+    mark = paddle.layer.data(name='mark_data', type=d_type(mark_dict_len))
+
+    target = paddle.layer.data(name='target', type=d_type(label_dict_len))
+
+    default_std = 1 / math.sqrt(hidden_dim) / 3.0
+
+    emb_para = paddle.attr.Param(name='emb', initial_std=0., learning_rate=0.)
+    std_0 = paddle.attr.Param(initial_std=0.)
+    std_default = paddle.attr.Param(initial_std=default_std)
+
+    predicate_embedding = paddle.layer.embedding(
+        size=word_dim,
+        input=predicate,
+        param_attr=paddle.attr.Param(
+            name='vemb', initial_std=default_std))
+    mark_embedding = paddle.layer.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 = [
+        paddle.layer.embedding(
+            size=word_dim, input=x, param_attr=emb_para) for x in word_input
+    ]
+    emb_layers.append(predicate_embedding)
+    emb_layers.append(mark_embedding)
+
+    hidden_0 = paddle.layer.mixed(
+        size=hidden_dim,
+        bias_attr=std_default,
+        input=[
+            paddle.layer.full_matrix_projection(
+                input=emb, param_attr=std_default) for emb in emb_layers
+        ])
+
+    mix_hidden_lr = 1e-3
+    lstm_para_attr = paddle.attr.Param(initial_std=0.0, learning_rate=1.0)
+    hidden_para_attr = paddle.attr.Param(
+        initial_std=default_std, learning_rate=mix_hidden_lr)
+
+    lstm_0 = paddle.layer.lstmemory(
+        input=hidden_0,
+        act=paddle.activation.Relu(),
+        gate_act=paddle.activation.Sigmoid(),
+        state_act=paddle.activation.Sigmoid(),
+        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 = paddle.layer.mixed(
+            size=hidden_dim,
+            bias_attr=std_default,
+            input=[
+                paddle.layer.full_matrix_projection(
+                    input=input_tmp[0], param_attr=hidden_para_attr),
+                paddle.layer.full_matrix_projection(
+                    input=input_tmp[1], param_attr=lstm_para_attr)
+            ])
+
+        lstm = paddle.layer.lstmemory(
+            input=mix_hidden,
+            act=paddle.activation.Relu(),
+            gate_act=paddle.activation.Sigmoid(),
+            state_act=paddle.activation.Sigmoid(),
+            reverse=((i % 2) == 1),
+            bias_attr=std_0,
+            param_attr=lstm_para_attr)
+
+        input_tmp = [mix_hidden, lstm]
+
+    feature_out = paddle.layer.mixed(
+        size=label_dict_len,
+        bias_attr=std_default,
+        input=[
+            paddle.layer.full_matrix_projection(
+                input=input_tmp[0], param_attr=hidden_para_attr),
+            paddle.layer.full_matrix_projection(
+                input=input_tmp[1], param_attr=lstm_para_attr)
+        ], )
+
+    crf_cost = paddle.layer.crf(size=label_dict_len,
+                                input=feature_out,
+                                label=target,
+                                param_attr=paddle.attr.Param(
+                                    name='crfw',
+                                    initial_std=default_std,
+                                    learning_rate=mix_hidden_lr))
+
+    crf_dec = paddle.layer.crf_decoding(
+        name='crf_dec_l',
+        size=label_dict_len,
+        input=feature_out,
+        label=target,
+        param_attr=paddle.attr.Param(name='crfw'))
+
+    return crf_cost, crf_dec
+
+
+def load_parameter(file_name, h, w):
+    with open(file_name, 'rb') as f:
+        f.read(16)  # skip header.
+        return np.fromfile(f, dtype=np.float32).reshape(h, w)
+
+
+def main():
+    paddle.init(use_gpu=False, trainer_count=1)
+
+    # define network topology
+    crf_cost, crf_dec = db_lstm()
+
+    # create parameters
+    parameters = paddle.parameters.create([crf_cost, crf_dec])
+
+    # create optimizer
+    optimizer = paddle.optimizer.Momentum(
+        momentum=0,
+        learning_rate=2e-2,
+        regularization=paddle.optimizer.L2Regularization(rate=8e-4),
+        model_average=paddle.optimizer.ModelAverage(
+            average_window=0.5, max_average_window=10000), )
+
+    def event_handler(event):
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 100 == 0:
+                print "Pass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics)
+
+    trainer = paddle.trainer.SGD(cost=crf_cost,
+                                 parameters=parameters,
+                                 update_equation=optimizer)
+    parameters.set('emb', load_parameter(conll05.get_embedding(), 44068, 32))
+
+    trn_reader = paddle.reader.batched(
+        paddle.reader.shuffle(
+            conll05.test(), buf_size=8192), batch_size=10)
+
+    reader_dict = {
+        'word_data': 0,
+        'ctx_n2_data': 1,
+        'ctx_n1_data': 2,
+        'ctx_0_data': 3,
+        'ctx_p1_data': 4,
+        'ctx_p2_data': 5,
+        'verb_data': 6,
+        'mark_data': 7,
+        'target': 8
+    }
+
+    trainer.train(
+        reader=trn_reader,
+        event_handler=event_handler,
+        num_passes=10000,
+        reader_dict=reader_dict)
+
+
+if __name__ == '__main__':
+    main()

demo/sentiment/train_v2.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+import sys
+import paddle.trainer_config_helpers.attrs as attrs
+from paddle.trainer_config_helpers.poolings import MaxPooling
+import paddle.v2 as paddle
+
+
+def convolution_net(input_dim,
+                    class_dim=2,
+                    emb_dim=128,
+                    hid_dim=128,
+                    is_predict=False):
+    data = paddle.layer.data("word",
+                             paddle.data_type.integer_value_sequence(input_dim))
+    emb = paddle.layer.embedding(input=data, size=emb_dim)
+    conv_3 = paddle.networks.sequence_conv_pool(
+        input=emb, context_len=3, hidden_size=hid_dim)
+    conv_4 = paddle.networks.sequence_conv_pool(
+        input=emb, context_len=4, hidden_size=hid_dim)
+    output = paddle.layer.fc(input=[conv_3, conv_4],
+                             size=class_dim,
+                             act=paddle.activation.Softmax())
+    lbl = paddle.layer.data("label", paddle.data_type.integer_value(2))
+    cost = paddle.layer.classification_cost(input=output, label=lbl)
+    return cost
+
+
+def stacked_lstm_net(input_dim,
+                     class_dim=2,
+                     emb_dim=128,
+                     hid_dim=512,
+                     stacked_num=3,
+                     is_predict=False):
+    """
+    A Wrapper for sentiment classification task.
+    This network uses bi-directional recurrent network,
+    consisting three LSTM layers. This configure is referred to
+    the paper as following url, but use fewer layrs.
+        http://www.aclweb.org/anthology/P15-1109
+
+    input_dim: here is word dictionary dimension.
+    class_dim: number of categories.
+    emb_dim: dimension of word embedding.
+    hid_dim: dimension of hidden layer.
+    stacked_num: number of stacked lstm-hidden layer.
+    is_predict: is predicting or not.
+                Some layers is not needed in network when predicting.
+    """
+    assert stacked_num % 2 == 1
+
+    layer_attr = attrs.ExtraLayerAttribute(drop_rate=0.5)
+    fc_para_attr = attrs.ParameterAttribute(learning_rate=1e-3)
+    lstm_para_attr = attrs.ParameterAttribute(initial_std=0., learning_rate=1.)
+    para_attr = [fc_para_attr, lstm_para_attr]
+    bias_attr = attrs.ParameterAttribute(initial_std=0., l2_rate=0.)
+    relu = paddle.activation.Relu()
+    linear = paddle.activation.Linear()
+
+    data = paddle.layer.data("word",
+                             paddle.data_type.integer_value_sequence(input_dim))
+    emb = paddle.layer.embedding(input=data, size=emb_dim)
+
+    fc1 = paddle.layer.fc(input=emb,
+                          size=hid_dim,
+                          act=linear,
+                          bias_attr=bias_attr)
+    lstm1 = paddle.layer.lstmemory(
+        input=fc1, act=relu, bias_attr=bias_attr, layer_attr=layer_attr)
+
+    inputs = [fc1, lstm1]
+    for i in range(2, stacked_num + 1):
+        fc = paddle.layer.fc(input=inputs,
+                             size=hid_dim,
+                             act=linear,
+                             param_attr=para_attr,
+                             bias_attr=bias_attr)
+        lstm = paddle.layer.lstmemory(
+            input=fc,
+            reverse=(i % 2) == 0,
+            act=relu,
+            bias_attr=bias_attr,
+            layer_attr=layer_attr)
+        inputs = [fc, lstm]
+
+    fc_last = paddle.layer.pooling(input=inputs[0], pooling_type=MaxPooling())
+    lstm_last = paddle.layer.pooling(input=inputs[1], pooling_type=MaxPooling())
+    output = paddle.layer.fc(input=[fc_last, lstm_last],
+                             size=class_dim,
+                             act=paddle.activation.Softmax(),
+                             bias_attr=bias_attr,
+                             param_attr=para_attr)
+
+    lbl = paddle.layer.data("label", paddle.data_type.integer_value(2))
+    cost = paddle.layer.classification_cost(input=output, label=lbl)
+    return cost
+
+
+if __name__ == '__main__':
+    # init
+    paddle.init(use_gpu=True, trainer_count=4)
+
+    # network config
+    print 'load dictionary...'
+    word_dict = paddle.dataset.imdb.word_dict()
+    dict_dim = len(word_dict)
+    class_dim = 2
+
+    # Please choose the way to build the network
+    # by uncommenting the corresponding line.
+    cost = convolution_net(dict_dim, class_dim=class_dim)
+    # cost = stacked_lstm_net(dict_dim, class_dim=class_dim, stacked_num=3)
+
+    # create parameters
+    parameters = paddle.parameters.create(cost)
+
+    # create optimizer
+    adam_optimizer = paddle.optimizer.Adam(
+        learning_rate=2e-3,
+        regularization=paddle.optimizer.L2Regularization(rate=8e-4),
+        model_average=paddle.optimizer.ModelAverage(average_window=0.5))
+
+    # End batch and end pass event handler
+    def event_handler(event):
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 100 == 0:
+                print "\nPass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics)
+            else:
+                sys.stdout.write('.')
+                sys.stdout.flush()
+        if isinstance(event, paddle.event.EndPass):
+            result = trainer.test(
+                reader=paddle.reader.batched(
+                    lambda: paddle.dataset.imdb.test(word_dict),
+                    batch_size=128),
+                reader_dict={'word': 0,
+                             'label': 1})
+            print "\nTest with Pass %d, %s" % (event.pass_id, result.metrics)
+
+    # create trainer
+    trainer = paddle.trainer.SGD(cost=cost,
+                                 parameters=parameters,
+                                 update_equation=adam_optimizer)
+
+    trainer.train(
+        reader=paddle.reader.batched(
+            paddle.reader.shuffle(
+                lambda: paddle.dataset.imdb.train(word_dict), buf_size=1000),
+            batch_size=100),
+        event_handler=event_handler,
+        reader_dict={'word': 0,
+                     'label': 1},
+        num_passes=10)

demo/seqToseq/api_train_v2.py

+import os
+
+import paddle.v2 as paddle
+
+from seqToseq_net_v2 import seqToseq_net_v2
+
+# Data Definiation.
+# TODO:This code should be merged to dataset package.
+data_dir = "./data/pre-wmt14"
+src_lang_dict = os.path.join(data_dir, 'src.dict')
+trg_lang_dict = os.path.join(data_dir, 'trg.dict')
+
+source_dict_dim = len(open(src_lang_dict, "r").readlines())
+target_dict_dim = len(open(trg_lang_dict, "r").readlines())
+
+
+def read_to_dict(dict_path):
+    with open(dict_path, "r") as fin:
+        out_dict = {
+            line.strip(): line_count
+            for line_count, line in enumerate(fin)
+        }
+    return out_dict
+
+
+src_dict = read_to_dict(src_lang_dict)
+trg_dict = read_to_dict(trg_lang_dict)
+
+train_list = os.path.join(data_dir, 'train.list')
+test_list = os.path.join(data_dir, 'test.list')
+
+UNK_IDX = 2
+START = "<s>"
+END = "<e>"
+
+
+def _get_ids(s, dictionary):
+    words = s.strip().split()
+    return [dictionary[START]] + \
+           [dictionary.get(w, UNK_IDX) for w in words] + \
+           [dictionary[END]]
+
+
+def train_reader(file_name):
+    def reader():
+        with open(file_name, 'r') as f:
+            for line_count, line in enumerate(f):
+                line_split = line.strip().split('\t')
+                if len(line_split) != 2:
+                    continue
+                src_seq = line_split[0]  # one source sequence
+                src_ids = _get_ids(src_seq, src_dict)
+
+                trg_seq = line_split[1]  # one target sequence
+                trg_words = trg_seq.split()
+                trg_ids = [trg_dict.get(w, UNK_IDX) for w in trg_words]
+
+                # remove sequence whose length > 80 in training mode
+                if len(src_ids) > 80 or len(trg_ids) > 80:
+                    continue
+                trg_ids_next = trg_ids + [trg_dict[END]]
+                trg_ids = [trg_dict[START]] + trg_ids
+
+                yield src_ids, trg_ids, trg_ids_next
+
+    return reader
+
+
+def main():
+    paddle.init(use_gpu=False, trainer_count=1)
+
+    # define network topology
+    cost = seqToseq_net_v2(source_dict_dim, target_dict_dim)
+    parameters = paddle.parameters.create(cost)
+    optimizer = paddle.optimizer.Adam(learning_rate=1e-4)
+
+    def event_handler(event):
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 10 == 0:
+                print "Pass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics)
+
+    trainer = paddle.trainer.SGD(cost=cost,
+                                 parameters=parameters,
+                                 update_equation=optimizer)
+
+    reader_dict = {
+        'source_language_word': 0,
+        'target_language_word': 1,
+        'target_language_next_word': 2
+    }
+
+    trn_reader = paddle.reader.batched(
+        paddle.reader.shuffle(
+            train_reader("data/pre-wmt14/train/train"), buf_size=8192),
+        batch_size=5)
+
+    trainer.train(
+        reader=trn_reader,
+        event_handler=event_handler,
+        num_passes=10000,
+        reader_dict=reader_dict)
+
+
+if __name__ == '__main__':
+    main()

demo/seqToseq/seqToseq_net_v2.py

+import paddle.v2.activation as activation
+import paddle.v2.attr as attr
+import paddle.v2.data_type as data_type
+import paddle.v2.layer as layer
+import paddle.v2.networks as networks
+
+
+def seqToseq_net_v2(source_dict_dim, target_dict_dim):
+    ### Network Architecture
+    word_vector_dim = 512  # dimension of word vector
+    decoder_size = 512  # dimension of hidden unit in GRU Decoder network
+    encoder_size = 512  # dimension of hidden unit in GRU Encoder network
+
+    #### Encoder
+    src_word_id = layer.data(
+        name='source_language_word',
+        type=data_type.integer_value_sequence(source_dict_dim))
+    src_embedding = layer.embedding(
+        input=src_word_id,
+        size=word_vector_dim,
+        param_attr=attr.ParamAttr(name='_source_language_embedding'))
+    src_forward = networks.simple_gru(input=src_embedding, size=encoder_size)
+    src_backward = networks.simple_gru(
+        input=src_embedding, size=encoder_size, reverse=True)
+    encoded_vector = layer.concat(input=[src_forward, src_backward])
+
+    #### Decoder
+    with layer.mixed(size=decoder_size) as encoded_proj:
+        encoded_proj += layer.full_matrix_projection(input=encoded_vector)
+
+    backward_first = layer.first_seq(input=src_backward)
+
+    with layer.mixed(size=decoder_size, act=activation.Tanh()) as decoder_boot:
+        decoder_boot += layer.full_matrix_projection(input=backward_first)
+
+    def gru_decoder_with_attention(enc_vec, enc_proj, current_word):
+
+        decoder_mem = layer.memory(
+            name='gru_decoder', size=decoder_size, boot_layer=decoder_boot)
+
+        context = networks.simple_attention(
+            encoded_sequence=enc_vec,
+            encoded_proj=enc_proj,
+            decoder_state=decoder_mem)
+
+        with layer.mixed(size=decoder_size * 3) as decoder_inputs:
+            decoder_inputs += layer.full_matrix_projection(input=context)
+            decoder_inputs += layer.full_matrix_projection(input=current_word)
+
+        gru_step = layer.gru_step(
+            name='gru_decoder',
+            input=decoder_inputs,
+            output_mem=decoder_mem,
+            size=decoder_size)
+
+        with layer.mixed(
+                size=target_dict_dim, bias_attr=True,
+                act=activation.Softmax()) as out:
+            out += layer.full_matrix_projection(input=gru_step)
+        return out
+
+    decoder_group_name = "decoder_group"
+    group_input1 = layer.StaticInputV2(input=encoded_vector, is_seq=True)
+    group_input2 = layer.StaticInputV2(input=encoded_proj, is_seq=True)
+    group_inputs = [group_input1, group_input2]
+
+    trg_embedding = layer.embedding(
+        input=layer.data(
+            name='target_language_word',
+            type=data_type.integer_value_sequence(target_dict_dim)),
+        size=word_vector_dim,
+        param_attr=attr.ParamAttr(name='_target_language_embedding'))
+    group_inputs.append(trg_embedding)
+
+    # For decoder equipped with attention mechanism, in training,
+    # target embeding (the groudtruth) is the data input,
+    # while encoded source sequence is accessed to as an unbounded memory.
+    # Here, the StaticInput defines a read-only memory
+    # for the recurrent_group.
+    decoder = layer.recurrent_group(
+        name=decoder_group_name,
+        step=gru_decoder_with_attention,
+        input=group_inputs)
+
+    lbl = layer.data(
+        name='target_language_next_word',
+        type=data_type.integer_value_sequence(target_dict_dim))
+    cost = layer.classification_cost(input=decoder, label=lbl)
+
+    return cost

doc/api/index_cn.rst

-API中文手册
-============
-
+API
+===
\ No newline at end of file
-DataProvider API
-----------------
-
-..  toctree::
-    :maxdepth: 1
-
-    data_provider/dataprovider_cn.rst
-    data_provider/pydataprovider2_cn.rst
-
-..  _api_trainer_config:
-
-Model Config API
-----------------
-
-..  toctree::
-    :maxdepth: 1
-
-    trainer_config_helpers/optimizers.rst
-    trainer_config_helpers/data_sources.rst
-    trainer_config_helpers/layers.rst
-    trainer_config_helpers/activations.rst 
-    trainer_config_helpers/poolings.rst
-    trainer_config_helpers/networks.rst
-    trainer_config_helpers/evaluators.rst
-    trainer_config_helpers/attrs.rst
-
-
-Applications API
-----------------
-
-..  toctree::
-    :maxdepth: 1
-
-    predict/swig_py_paddle_cn.rst

doc/api/index_en.rst

 API
 ===
 
-DataProvider API
-----------------
-
-..  toctree::
-    :maxdepth: 1
-
-    data_provider/dataprovider_en.rst
-    data_provider/pydataprovider2_en.rst
-
-..  _api_trainer_config:
-
 Model Config API
 ----------------
 
 ..  toctree::
     :maxdepth: 1
 
-    trainer_config_helpers/optimizers.rst
-    trainer_config_helpers/data_sources.rst
+    v2/model_configs.rst
\ No newline at end of file
-    trainer_config_helpers/layers.rst
-    trainer_config_helpers/activations.rst 
-    trainer_config_helpers/poolings.rst
-    trainer_config_helpers/networks.rst
-    trainer_config_helpers/evaluators.rst
-    trainer_config_helpers/attrs.rst
-
-
-Applications API
-----------------
-
-..  toctree::
-    :maxdepth: 1
-
-    predict/swig_py_paddle_en.rst

doc/api/v1/index_cn.rst

+API中文手册
+============
+
+DataProvider API
+----------------
+
+..  toctree::
+    :maxdepth: 1
+
+    data_provider/dataprovider_cn.rst
+    data_provider/pydataprovider2_cn.rst
+
+..  _api_trainer_config:
+
+Model Config API
+----------------
+
+..  toctree::
+    :maxdepth: 1
+
+    trainer_config_helpers/optimizers.rst
+    trainer_config_helpers/data_sources.rst
+    trainer_config_helpers/layers.rst
+    trainer_config_helpers/activations.rst 
+    trainer_config_helpers/poolings.rst
+    trainer_config_helpers/networks.rst
+    trainer_config_helpers/evaluators.rst
+    trainer_config_helpers/attrs.rst
+
+
+Applications API
+----------------
+
+..  toctree::
+    :maxdepth: 1
+
+    predict/swig_py_paddle_cn.rst

doc/api/v1/index_en.rst

+API
+===
+
+DataProvider API
+----------------
+
+..  toctree::
+    :maxdepth: 1
+
+    data_provider/dataprovider_en.rst
+    data_provider/pydataprovider2_en.rst
+
+..  _api_trainer_config:
+
+Model Config API
+----------------
+
+..  toctree::
+    :maxdepth: 1
+
+    trainer_config_helpers/optimizers.rst
+    trainer_config_helpers/data_sources.rst
+    trainer_config_helpers/layers.rst
+    trainer_config_helpers/activations.rst 
+    trainer_config_helpers/poolings.rst
+    trainer_config_helpers/networks.rst
+    trainer_config_helpers/evaluators.rst
+    trainer_config_helpers/attrs.rst
+
+
+Applications API
+----------------
+
+..  toctree::
+    :maxdepth: 1
+
+    predict/swig_py_paddle_en.rst

doc/api/v2/model_configs.rst

+======
+Layers
+======
+
+..  automodule:: paddle.v2.layer
+    :members:

doc/design/reader/README.md

@@ -4,9 +4,10 @@ At training and testing time, PaddlePaddle programs need to read data. To ease t
 
 - A *reader* is a function that reads data (from file, network, random number generator, etc) and yields data items.
 - A *reader creator* is a function that returns a reader function.
-- A *reader* decorator is a function, which accepts one or more readers, and returns a reader.
+- A *reader decorator* is a function, which accepts one or more readers, and returns a reader.
+- A *batch reader* is a function that reads data (from *reader*, file, network, random number generator, etc) and yields a batch of data items.
 
-and provide frequently used reader creators and reader decorators.
+and provide function which converts reader to batch reader, frequently used reader creators and reader decorators.
 
 ## Data Reader Interface
 
@@ -37,9 +38,54 @@ def reader_creator_random_imageand_label(widht, height, label):
 	return reader
 ```
 
+## Batch Reader Interface
+
+*batch reader* can be any function with no parameter that creates a iterable (anything can be used in `for x in iterable`). The output of the iterable should be a batch (list) of data items. Each item inside the list must be a tuple.
+
+Here are valid outputs:
+```python
+# a mini batch of three data items. Each data item consist three columns of data, each of which is 1.
+[(1, 1, 1),
+(2, 2, 2),
+(3, 3, 3)]
+
+# a mini batch of three data items, each data item is a list (single column).
+[([1,1,1],),
+([2,2,2],),
+([3,3,3],),
+```
+
+Please note that each item inside the list must be a tuple, below is an invalid output:
+```python
+ # wrong, [1,1,1] needs to be inside a tuple: ([1,1,1],).
+ # Otherwise it's ambiguous whether [1,1,1] means a single column of data [1, 1, 1],
+ # or three column of datas, each of which is 1.
+[[1,1,1],
+[2,2,2],
+[3,3,3]]
+```
+
+It's easy to convert from reader to batch reader:
+```python
+mnist_train = paddle.dataset.mnist.train()
+mnist_train_batch_reader = paddle.batch(mnist_train, 128)
+```
+
+Also easy to create custom batch reader:
+```python
+def custom_batch_reader():
+	while True:
+		batch = []
+		for i in xrange(128):
+			batch.append((numpy.random.uniform(-1, 1, 28*28),)) # note that it's a tuple being appended.
+		yield batch
+
+mnist_random_image_batch_reader = custom_batch_reader
+```
+
 ## Usage
 
-data reader, mapping from item(s) read to data layer, batch size and number of total pass will be passed into `paddle.train`:
+batch reader, mapping from item(s) read to data layer, batch size and number of total pass will be passed into `paddle.train`:
 
 ```python
 # two data layer is created:
@@ -47,8 +93,8 @@ image_layer = paddle.layer.data("image", ...)
 label_layer = paddle.layer.data("label", ...)
 
 # ...
-
-paddle.train(paddle.dataset.mnist, {"image":0, "label":1}, 128, 10, ...)
+batch_reader = paddle.batch(paddle.dataset.mnist.train(), 128)
+paddle.train(batch_reader, {"image":0, "label":1}, 128, 10, ...)
 ```
 
 ## Data Reader Decorator
@@ -64,7 +110,7 @@ Since reading data may take time and training can not proceed without data. It i
 Use `paddle.reader.buffered` to prefetch data:
 
 ```python
-buffered_reader = paddle.reader.buffered(paddle.dataset.mnist, 100)
+buffered_reader = paddle.reader.buffered(paddle.dataset.mnist.train(), 100)
 ```
 
 `buffered_reader` will try to buffer (prefetch) `100` data entries.
@@ -91,10 +137,10 @@ def reader_creator_bool(t):
 true_reader = reader_creator_bool(True)
 false_reader = reader_creator_bool(False)
 
-reader = paddle.reader.compose(paddle.dataset.mnist, data_reader_creator_random_image(20, 20), true_reader, false_reader)
-# Skipped 1 because paddle.dataset.mnist produces two items per data entry.
+reader = paddle.reader.compose(paddle.dataset.mnist.train(), data_reader_creator_random_image(20, 20), true_reader, false_reader)
+# Skipped 1 because paddle.dataset.mnist.train() produces two items per data entry.
 # And we don't care second item at this time.
-paddle.train(reader, {"true_image":0, "fake_image": 2, "true_label": 3, "false_label": 4}, ...)
+paddle.train(paddle.batch(reader, 128), {"true_image":0, "fake_image": 2, "true_label": 3, "false_label": 4}, ...)
 ```
 
 ### Shuffle
@@ -103,16 +149,20 @@ Given shuffle buffer size `n`, `paddle.reader.shuffle` will return a data reader
 
 Example:
 ```python
-reader = paddle.reader.shuffle(paddle.dataset.mnist, 512)
+reader = paddle.reader.shuffle(paddle.dataset.mnist.train(), 512)
 ```
 
 ## Q & A
 
-### Why return only a single entry, but not a mini batch?
+### Why reader return only a single entry, but not a mini batch?
+
+Always returning a single entry make reusing existing data readers much easier (e.g., if existing reader return not a single entry but 3 entries, training code will be more complex because it need to handle cases like batch size 2).
+
+We provide function `paddle.batch` to turn (single entry) reader into batch reader.
 
-If a mini batch is returned, data reader need to take care of batch size. But batch size is a concept for training, it makes more sense for user to specify batch size as a parameter for `train`.
+### Why do we need batch reader, isn't train take reader and batch_size as arguments sufficient?
 
-Practically, always return a single entry make reusing existing data readers much easier (e.g., if existing reader return not a single entry but 3 entries, training code will be more complex because it need to handle cases like batch size 2).
+In most of the case, train taking reader and batch_size as arguments would be sufficent. However sometimes user want to customize order of data entries inside a mini batch. Or even change batch size dynamically.
 
 ### Why use a dictionary but not a list to provide mapping?
 
@@ -137,7 +187,7 @@ def image_reader_creator(image_path, label_path, n):
 
 # images_reader_creator creates a reader
 reader = image_reader_creator("/path/to/image_file", "/path/to/label_file", 1024)
-paddle.train(reader, {"image":0, "label":1}, ...)
+paddle.train(paddle.batch(reader, 128), {"image":0, "label":1}, ...)
 ```
 
 ### How is `paddle.train` implemented
@@ -145,17 +195,8 @@ paddle.train(reader, {"image":0, "label":1}, ...)
 An example implementation of paddle.train could be:
 
 ```python
-def make_minibatch(reader, minibatch_size):
-	def ret():
-		r = reader()
-		buf = [r.next() for x in xrange(minibatch_size)]
-		while len(buf) > 0:
-			yield buf
-			buf = [r.next() for x in xrange(minibatch_size)]
-	return ret
-
-def train(reader, mapping, batch_size, total_pass):
+def train(batch_reader, mapping, batch_size, total_pass):
 	for pass_idx in range(total_pass):
-		for mini_batch in make_minibatch(reader): # this loop will never end in online learning.
+		for mini_batch in batch_reader(): # this loop will never end in online learning.
 			do_forward_backward(mini_batch, mapping)
 ```

doc/howto/usage/k8s/src/k8s_train/start_paddle.py

@@ -132,7 +132,8 @@ def startPaddle(idMap={}, train_args_dict=None):
     logDir = JOB_PATH_OUTPUT + "/node_" + str(trainerId)
     if not os.path.exists(JOB_PATH_OUTPUT):
         os.makedirs(JOB_PATH_OUTPUT)
-    os.mkdir(logDir)
+    if not os.path.exists(logDir):
+        os.mkdir(logDir)
     copyCommand = 'cp -rf ' + JOB_PATH + \
         "/" + str(trainerId) + "/data/*" + " ./data/"
     os.system(copyCommand)

doc/templates/conf.py.cn.in

@@ -15,13 +15,19 @@ import sys
 import os, subprocess
 import shlex
 from recommonmark import parser, transform
+try:
+   import py_paddle
+   import paddle
+   import paddle.v2
+except ImportError:
+   print("Must install paddle python package before generating documentation")
+   sys.exit(1)
 
 MarkdownParser = parser.CommonMarkParser
 AutoStructify = transform.AutoStructify
 # If extensions (or modules to document with autodoc) are in another directory,
 # add these directories to sys.path here. If the directory is relative to the
 # documentation root, use os.path.abspath to make it absolute, like shown here.
-sys.path.insert(0, '@PROJ_ROOT@/python')
 templates_path = ["@PROJ_ROOT@/doc_theme/templates"]
 
 # -- General configuration ------------------------------------------------

doc/templates/conf.py.en.in

@@ -15,14 +15,20 @@ import sys
 import os, subprocess
 import shlex
 from recommonmark import parser, transform
+try:
+   import py_paddle
+   import paddle
+   import paddle.v2
+except ImportError:
+   print("Must install paddle python package before generating documentation")
+   sys.exit(1)
+
 
 MarkdownParser = parser.CommonMarkParser
 AutoStructify = transform.AutoStructify
 # If extensions (or modules to document with autodoc) are in another directory,
 # add these directories to sys.path here. If the directory is relative to the
 # documentation root, use os.path.abspath to make it absolute, like shown here.
-sys.path.insert(0, '@PROJ_ROOT@/python')
-
 templates_path = ["@PROJ_ROOT@/doc_theme/templates"]
 
 # -- General configuration ------------------------------------------------

doc/tutorials/quick_start/index_en.md

@@ -156,14 +156,14 @@ define_py_data_sources2(train_list='data/train.list',
                         obj="process",
                         args={"dictionary": word_dict})
 ```
-You can refer to the following link for more detailed examples and data formats: <a href = "../../api/data_provider/pydataprovider2_en.html">PyDataProvider2</a>.
+You can refer to the following link for more detailed examples and data formats: <a href = "../../api/v1/data_provider/pydataprovider2_en.html">PyDataProvider2</a>.
 
 ## Network Architecture
 We will describe four kinds of network architectures in this section.
 <center> ![](./src/PipelineNetwork_en.jpg) </center>
 
 First, you will build a logistic regression model. Later, you will also get chance to build other more powerful network architectures.
-For more detailed documentation, you could refer to: <a href = "../../api/trainer_config_helpers/layers.html">layer documentation</a>. All configuration files are in `demo/quick_start` directory.
+For more detailed documentation, you could refer to: <a href = "../../api/v1/trainer_config_helpers/layers.html">layer documentation</a>. All configuration files are in `demo/quick_start` directory.
 
 ### Logistic Regression
 The architecture is illustrated in the following picture:
@@ -366,7 +366,7 @@ You can use single layer LSTM model with Dropout for our text classification pro
 <br>
 
 ## Optimization Algorithm
-<a href = "../../api/trainer_config_helpers/optimizers.html">Optimization algorithms</a> include Momentum, RMSProp, AdaDelta, AdaGrad, Adam, and Adamax. You can use Adam optimization method here, with L2 regularization and gradient clipping, because Adam has been proved to work very well for training recurrent neural network.
+<a href = "../../api/v1/trainer_config_helpers/optimizers.html">Optimization algorithms</a> include Momentum, RMSProp, AdaDelta, AdaGrad, Adam, and Adamax. You can use Adam optimization method here, with L2 regularization and gradient clipping, because Adam has been proved to work very well for training recurrent neural network.
 
 ```python
 settings(batch_size=128,
@@ -407,7 +407,7 @@ paddle train \
 --init_model_path=./output/pass-0000x
 ```
 
-We will give an example of performing prediction using Recurrent model on a dataset with no labels. You can refer to <a href = "../../api/predict/swig_py_paddle_en.html">Python Prediction API</a> tutorial，or other <a href = "../../tutorials/index_en.html">demo</a> for the prediction process using Python. You can also use the following script for inference or evaluation.
+We will give an example of performing prediction using Recurrent model on a dataset with no labels. You can refer to <a href = "../../api/v1/predict/swig_py_paddle_en.html">Python Prediction API</a> tutorial，or other <a href = "../../tutorials/index_en.html">demo</a> for the prediction process using Python. You can also use the following script for inference or evaluation.
 
 inference script (predict.sh)：
 

paddle/api/Arguments.cpp

@@ -144,9 +144,7 @@ void Arguments::setSlotSequenceDim(size_t idx, IVector* vec) throw(RangeError) {
   a.cpuSequenceDims = m->cast<paddle::IVector>(vec->getSharedPtr());
 }
 
-float Arguments::sumCosts() const {
-  return paddle::Argument::sumCosts(m->outputs);
-}
+float Arguments::sum() const { return paddle::Argument::sum(m->outputs); }
 
 int64_t Arguments::getBatchSize(size_t idx) const throw(RangeError) {
   auto& a = m->getArg(idx);

paddle/api/GradientMachine.cpp

@@ -142,6 +142,20 @@ Parameter* GradientMachine::getParameter(size_t i) throw(RangeError) {
   }
 }
 
+size_t GradientMachine::getNonStaticParameterSize() const {
+  return m->machine->getNonStaticParameters().size();
+}
+
+Parameter* GradientMachine::getNonStaticParameter(size_t i) throw(RangeError) {
+  auto params = m->machine->getNonStaticParameters();
+  if (i < params.size()) {
+    return Parameter::createFromSharedPtr(
+        &m->machine->getNonStaticParameters()[i]);
+  } else {
+    throw RangeError();
+  }
+}
+
 void GradientMachine::randParameters() { m->machine->randParameters(); }
 
 Arguments* GradientMachine::getLayerOutput(const std::string& layerName) const

paddle/api/PaddleAPI.h

@@ -453,7 +453,7 @@ public:
                                         IVector* vec) throw(RangeError);
   void setSlotSequenceDim(size_t idx, IVector* vec) throw(RangeError);
 
-  float sumCosts() const;
+  float sum() const;
 
 private:
   static Arguments* createByPaddleArgumentVector(void* ptr);
@@ -771,6 +771,9 @@ public:
   size_t getParameterSize() const;
   Parameter* getParameter(size_t i) throw(RangeError);
 
+  size_t getNonStaticParameterSize() const;
+  Parameter* getNonStaticParameter(size_t i) throw(RangeError);
+
   void randParameters();
 
   Arguments* getLayerOutput(const std::string& layerName) const

paddle/api/test/testArguments.py

@@ -22,7 +22,7 @@ class TestArguments(unittest.TestCase):
         args = swig_paddle.Arguments.createArguments(1)
         args.setSlotValue(0, m)
 
-        self.assertAlmostEqual(27.0, args.sumCosts())
+        self.assertAlmostEqual(27.0, args.sum())
 
         mat = args.getSlotValue(0)
         assert isinstance(mat, swig_paddle.Matrix)

paddle/gserver/tests/LayerGradUtil.cpp

@@ -24,7 +24,7 @@ real getCostSum(LayerPtr& testLayer, MatrixPtr weights) {
   if (weights) {
     outArgs[0].value->dotMul(*outArgs[0].value, *weights);
   }
-  return Argument::sumCosts(outArgs);
+  return Argument::sum(outArgs);
 }
 
 real getDiffAndPrint(real newCost1,
@@ -241,7 +241,7 @@ void testBatchState(LayerPtr testLayer,
 
     std::vector<Argument> args;
     args.push_back(out);
-    EXPECT_EQ(0, Argument::sumCosts(args)) << "testBatchState failed";
+    EXPECT_EQ(0, Argument::sum(args)) << "testBatchState failed";
     for (size_t seqId = 0; seqId < numSequences; ++seqId) {
       start[seqId] += seqLens[seqId];
     }
@@ -672,7 +672,7 @@ void testLayerGradKernel(TestConfig testConf,
     outArgs[0].value->dotMul(*testLayer->getOutput().value, *weights);
   }
 
-  real cost = Argument::sumCosts(outArgs);
+  real cost = Argument::sum(outArgs);
   LOG(INFO) << " cost " << cost;
   EXPECT_FALSE(std::isnan(cost));
 

paddle/parameter/Argument.h

@@ -163,7 +163,7 @@ struct Argument {
                        : sequenceStartPositions->getData(false);
   }
 
-  static inline real sumCosts(const std::vector<Argument>& arguments) {
+  static inline real sum(const std::vector<Argument>& arguments) {
     real cost = 0;
     for (auto& arg : arguments) {
       if (arg.value) {

paddle/pserver/test/CMakeLists.txt

@@ -10,9 +10,11 @@ add_test(NAME socket_test
 add_unittest_without_exec(test_ProtoServer
     test_ProtoServer.cpp)
 
-add_test(NAME test_ProtoServer
-    COMMAND ${PROJ_ROOT}/paddle/.set_port.sh -p port
-        ${CMAKE_CURRENT_BINARY_DIR}/test_ProtoServer)
+IF(NOT ON_TRAVIS)
+    add_test(NAME test_ProtoServer
+        COMMAND ${PROJ_ROOT}/paddle/.set_port.sh -p port
+            ${CMAKE_CURRENT_BINARY_DIR}/test_ProtoServer)
+ENDIF(NOT ON_TRAVIS)
 
 # TODO(yuyang18): Run test_ProtoServer when with rdma
 # add_test(NAME test_ProtoServerRDMA

paddle/py_paddle/util.py

@@ -195,6 +195,12 @@ def __monkeypatch_gradient_machine__():
 
     swig_paddle.GradientMachine.getParameters = getParameters
 
+    def getNonStaticParameters(self):
+        return (self.getNonStaticParameter(i)
+                for i in xrange(self.getNonStaticParameterSize()))
+
+    swig_paddle.GradientMachine.getNonStaticParameters = getNonStaticParameters
+
     def getLayerOutputs(self, layerNames):
         """
         getLayerOutputs. get outputs of layers and return a numpy matrix dict.

paddle/scripts/travis/before_install.osx.sh

-#!/bin/bash
-brew update
-brew tap homebrew/science
-brew install openblas swig md5sha1sum

paddle/scripts/travis/build_and_test.sh

@@ -2,18 +2,11 @@
 source ./common.sh
 
 NPROC=1
-if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then
-  export PYTHONPATH=/opt/python/2.7.12/lib/python2.7/site-packages
-  export PYTHONHOME=/opt/python/2.7.12
-  export PATH=/opt/python/2.7.12/bin:${PATH}
-  cmake .. -DCMAKE_Fortran_COMPILER=/usr/bin/gfortran-4.8 -DON_TRAVIS=ON -DON_COVERALLS=ON -DCOVERALLS_UPLOAD=ON ${EXTRA_CMAKE_OPTS}
-  NRPOC=`nproc`
-  make -j $NPROC
-  make coveralls
-  sudo make install
-elif [[ "$TRAVIS_OS_NAME" == "osx" ]]; then
-  export PYTHONPATH=/usr/local/lib/python2.7/site-packages
-  cmake .. -DON_TRAVIS=ON -DON_COVERALLS=ON -DCOVERALLS_UPLOAD=ON ${EXTRA_CMAKE_OPTS}
-  NPROC=`sysctl -n hw.ncpu`
-  make -j $NPROC
-fi
+export PYTHONPATH=/opt/python/2.7.12/lib/python2.7/site-packages
+export PYTHONHOME=/opt/python/2.7.12
+export PATH=/opt/python/2.7.12/bin:${PATH}
+cmake .. -DCMAKE_Fortran_COMPILER=/usr/bin/gfortran-4.8 -DON_TRAVIS=ON -DON_COVERALLS=ON -DCOVERALLS_UPLOAD=ON ${EXTRA_CMAKE_OPTS}
+NRPOC=`nproc`
+make -j $NPROC
+make coveralls
+sudo make install

paddle/scripts/travis/docs.sh

@@ -2,8 +2,12 @@
 
 # Add set -e, cd to directory.
 source ./common.sh
-
 # Compile Documentation only.
+cmake .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_Fortran_COMPILER=/usr/bin/gfortran-4.8 -DWITH_GPU=OFF -DWITH_DOC=OFF -DWITH_STYLE_CHECK=OFF ${EXTRA_CMAKE_OPTS}
+mkdir output
+make DESTDIR=./output install -j `nproc`
+pip install ./output/usr/local/opt/paddle/share/wheels/*
+rm -rf *
 cmake .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_Fortran_COMPILER=/usr/bin/gfortran-4.8 -DWITH_GPU=OFF -DWITH_DOC=ON ${EXTRA_CMAKE_OPTS}
 make paddle_docs paddle_docs_cn
 
@@ -25,26 +29,41 @@ TARGET_BRANCH="gh-pages"
 # Only deploy master branch to build latest documentation.
 SOURCE_BRANCH="master"
 
-# If is not a Github pull request, and in master branch.
-if [ "$TRAVIS_PULL_REQUEST" != "false" -o "$TRAVIS_BRANCH" != "$SOURCE_BRANCH"  ]; then
-  exit 0
-fi
-
 # Clone the repo to output directory
 git clone $REPO output
 cd output
 
-# checkout github page branch
-git checkout $TARGET_BRANCH || git checkout --orphan $TARGET_BRANCH
+function deploy_docs() {
+  SOURCE_BRANCH=$1
+  DIR=$2
+  # If is not a Github pull request
+  if [ "$TRAVIS_PULL_REQUEST" != "false" ]; then
+    exit 0
+  fi
+  # If it is not watched branch.
+  if [ "$TRAVIS_BRANCH" != "$SOURCE_BRANCH" ]; then
+    return
+  fi
 
-# remove old docs. mv new docs.
-rm -rf doc doc_cn
-mv ../doc/cn/html doc_cn
-mv ../doc/en/html doc
+  # checkout github page branch
+  git checkout $TARGET_BRANCH || git checkout --orphan $TARGET_BRANCH
+  
+  mkdir -p ${DIR}
+  # remove old docs. mv new docs.
+  set +e
+  rm -rf ${DIR}/doc ${DIR}/doc_cn
+  set -e
+  mv ../doc/cn/html ${DIR}/doc_cn
+  mv ../doc/en/html ${DIR}/doc
+  git add .
+}
+
+deploy_docs "master" "." 
+deploy_docs "develop" "./develop/"
 
 # Check is there anything changed.
 set +e
-git diff --exit-code >/dev/null
+git diff --cached --exit-code >/dev/null
 if [ $? -eq 0 ]; then
   echo "No changes to the output on this push; exiting."
   exit 0
@@ -57,7 +76,6 @@ if [ -n $SSL_KEY ]; then  # Only push updated docs for github.com/PaddlePaddle/P
   git config user.name "Travis CI"
   git config user.email "paddle-dev@baidu.com"
   git commit -m "Deploy to GitHub Pages: ${SHA}"
-
   # Set ssh private key
   openssl aes-256-cbc -K $SSL_KEY -iv $SSL_IV -in ../../paddle/scripts/travis/deploy_key.enc -out deploy_key -d
   chmod 600 deploy_key

paddle/setup.py.in

@@ -72,6 +72,7 @@ setup(name="py_paddle",
   packages=['py_paddle'],
   include_dirs = include_dirs,
   install_requires = [
+    'nltk>=3.2.2',
     'numpy>=1.8.0',      # The numpy is required.
     'protobuf>=3.0.0'    # The paddle protobuf version
   ],

paddle/trainer/Tester.cpp

@@ -208,7 +208,7 @@ real Tester::forwardOneBatch(const DataBatch& dataBatch,
     return 0.0;  // In this case, there is no meaning to calculate cost
   }
 
-  return Argument::sumCosts(outArgs);
+  return Argument::sum(outArgs);
 }
 
 void Tester::testOnePassBatch(int passId) {

paddle/trainer/Trainer.cpp

@@ -310,7 +310,7 @@ real Trainer::checkGradient() {
   std::vector<Argument> outArgs;
 
   trainerInternal_.getGradientMachine()->forward(inArgs, &outArgs, PASS_GC);
-  real cost = Argument::sumCosts(outArgs);
+  real cost = Argument::sum(outArgs);
   LOG(INFO) << "original cost=" << cost;
   trainerInternal_.getGradientMachine()->backward();
 
@@ -340,7 +340,7 @@ real Trainer::checkGradient() {
     parameter->getBuf(PARAMETER_VALUE)->copyFrom(newPara);
     parameter->setValueUpdated();
     trainerInternal_.getGradientMachine()->forward(inArgs, &outArgs, PASS_GC);
-    real newCost1 = Argument::sumCosts(outArgs);
+    real newCost1 = Argument::sum(outArgs);
 
     for (size_t i = 0; i < dim; ++i) {
       newp[i] = oldp[i] - step * d[i];
@@ -349,7 +349,7 @@ real Trainer::checkGradient() {
     parameter->getBuf(PARAMETER_VALUE)->copyFrom(newPara);
     parameter->setValueUpdated();
     trainerInternal_.getGradientMachine()->forward(inArgs, &outArgs, PASS_GC);
-    real newCost2 = Argument::sumCosts(outArgs);
+    real newCost2 = Argument::sum(outArgs);
 
     real trueDelta = 0.5 * (newCost1 - newCost2);
     real diff = (1e-20 + trueDelta) / (1e-20 + delta) - 1;
@@ -575,7 +575,7 @@ real Trainer::calcGradient(const DataBatch& dataBatch,
 
   trainerInternal_.getGradientMachine()->forwardBackward(
       inArgs, &outArgs, PASS_TRAIN);
-  real cost = Argument::sumCosts(outArgs);
+  real cost = Argument::sum(outArgs);
 
   offset = 0;
   for (auto& para : parameters) {

paddle/trainer/TrainerInternal.cpp

@@ -134,7 +134,7 @@ void TrainerInternal::trainOneBatch(int64_t batchId,
   real cost = 0;
   {
     REGISTER_TIMER("sumCost");
-    cost = Argument::sumCosts(*outArgs);
+    cost = Argument::sum(*outArgs);
   }
 
   if (batchId % intconfig_->log_period == 0) {

python/paddle/trainer/PyDataProvider2.py

@@ -65,14 +65,18 @@ def sparse_value_slot(dim, seq_type=SequenceType.NO_SEQUENCE):
     return InputType(dim, seq_type, DataType.SparseValue)
 
 
-def index_slot(dim, seq_type=SequenceType.NO_SEQUENCE):
-    return InputType(dim, seq_type, DataType.Index)
+def index_slot(value_range, seq_type=SequenceType.NO_SEQUENCE):
+    """Data type of integer.
+    :param value_range: range of this integer.
+    """
+    return InputType(value_range, seq_type, DataType.Index)
 
 
 dense_vector = dense_slot
 sparse_binary_vector = sparse_non_value_slot
 sparse_vector = sparse_value_slot
 integer_value = index_slot
+integer_value.__doc__ = index_slot.__doc__
 
 
 def dense_vector_sequence(dim):
@@ -99,8 +103,11 @@ def sparse_vector_sub_sequence(dim):
     return sparse_vector(dim, seq_type=SequenceType.SUB_SEQUENCE)
 
 
-def integer_value_sequence(dim):
-    return integer_value(dim, seq_type=SequenceType.SEQUENCE)
+def integer_value_sequence(value_range):
+    """Data type of a sequence of integer.
+    :param value_range: range of each element.
+    """
+    return integer_value(value_range, seq_type=SequenceType.SEQUENCE)
 
 
 def integer_value_sub_sequence(dim):
@@ -108,6 +115,7 @@ def integer_value_sub_sequence(dim):
 
 
 integer_sequence = integer_value_sequence
+integer_sequence.__doc__ = integer_value_sequence.__doc__
 
 
 class SingleSlotWrapper(object):

python/paddle/v2/__init__.py

@@ -20,18 +20,19 @@ import event
 import data_type
 import topology
 import data_feeder
+import networks
 from . import dataset
 from . import reader
 import attr
 import pooling
-import inferencer
+import inference
 import networks
 import py_paddle.swig_paddle as api
 
 __all__ = [
     'optimizer', 'layer', 'activation', 'parameters', 'init', 'trainer',
     'event', 'data_type', 'attr', 'pooling', 'data_feeder', 'dataset', 'reader',
-    'topology', 'networks', 'inferencer', 'infer'
+    'topology', 'networks', 'infer'
 ]
 
 
@@ -43,4 +44,4 @@ def init(**kwargs):
     api.initPaddle(*args)
 
 
-infer = inferencer.infer
+infer = inference.infer

python/paddle/v2/config_base.py

@@ -22,6 +22,7 @@ class Layer(object):
     def __init__(self, name=None, parent_layers=None):
         assert isinstance(parent_layers, dict)
         self.name = name
+        self.__contex__ = {}
         self.__parent_layers__ = parent_layers
 
     def to_proto(self, context):
@@ -39,16 +40,38 @@ class Layer(object):
                                self.__parent_layers__[layer_name])
             kwargs[layer_name] = v1_layer
 
-        if self.name is None:
+        if self.context_name() is None:
             return self.to_proto_impl(**kwargs)
-        elif self.name not in context:
-            context[self.name] = self.to_proto_impl(**kwargs)
-
-        return context[self.name]
+        elif self.context_name() not in context:
+            context[self.context_name()] = self.to_proto_impl(**kwargs)
+        self.__contex__ = context
+        if self.use_context_name():
+            return context[self.context_name()]
+        else:
+            return context[self.name]
 
     def to_proto_impl(self, **kwargs):
         raise NotImplementedError()
 
+    def context_name(self):
+        """
+        Context name means the context which stores `to_proto_impl` result.
+        If multiple layer share same context_name, the `to_proto_impl` of them
+        will be invoked only once.
+        """
+        return self.name
+
+    def use_context_name(self):
+        return False
+
+    def calculate_size(self):
+        """
+        lazy calculate size of the layer, should be called when to_proto_impl of
+        this layer is called.
+        :return:
+        """
+        return self.__contex__[self.context_name()].size
+
 
 def __convert_to_v2__(method_name, parent_names, is_default_name=True):
     if is_default_name:

python/paddle/v2/dataset/__init__.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import mnist
+import imikolov
+import imdb
+import cifar
+import movielens
+import conll05
+import uci_housing
+import sentiment
+import wmt14
 
-__all__ = ['mnist']
+__all__ = [
+    'mnist', 'imikolov', 'imdb', 'cifar', 'movielens', 'conll05', 'sentiment'
+    'uci_housing', 'wmt14'
+]

python/paddle/v2/dataset/cifar.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
 """
 CIFAR dataset: https://www.cs.toronto.edu/~kriz/cifar.html
 """
+
 import cPickle
 import itertools
 import numpy

python/paddle/v2/dataset/common.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import requests
 import hashlib
 import os
 import shutil
+import sys
 
 __all__ = ['DATA_HOME', 'download', 'md5file']
 
@@ -27,9 +42,24 @@ def download(url, module_name, md5sum):
 
     filename = os.path.join(dirname, url.split('/')[-1])
     if not (os.path.exists(filename) and md5file(filename) == md5sum):
+        print "Cache file %s not found, downloading %s" % (filename, url)
         r = requests.get(url, stream=True)
-        with open(filename, 'w') as f:
-            shutil.copyfileobj(r.raw, f)
+        total_length = r.headers.get('content-length')
+
+        if total_length is None:
+            with open(filename, 'w') as f:
+                shutil.copyfileobj(r.raw, f)
+        else:
+            with open(filename, 'w') as f:
+                dl = 0
+                total_length = int(total_length)
+                for data in r.iter_content(chunk_size=4096):
+                    dl += len(data)
+                    f.write(data)
+                    done = int(50 * dl / total_length)
+                    sys.stdout.write("\r[%s%s]" % ('=' * done,
+                                                   ' ' * (50 - done)))
+                    sys.stdout.flush()
 
     return filename
 

python/paddle/v2/dataset/conll05.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+import tarfile
+import gzip
+import itertools
+from common import download
+
+__all__ = ['test, get_dict', 'get_embedding']
+"""
+Conll 2005 dataset.  Paddle semantic role labeling Book and demo use this
+dataset as an example. Because Conll 2005 is not free in public, the default
+downloaded URL is test set of Conll 2005 (which is public). Users can change
+URL and MD5 to their Conll dataset.
+"""
+
+DATA_URL = 'http://www.cs.upc.edu/~srlconll/conll05st-tests.tar.gz'
+DATA_MD5 = '387719152ae52d60422c016e92a742fc'
+WORDDICT_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/wordDict.txt'
+WORDDICT_MD5 = 'ea7fb7d4c75cc6254716f0177a506baa'
+VERBDICT_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/verbDict.txt'
+VERBDICT_MD5 = '0d2977293bbb6cbefab5b0f97db1e77c'
+TRGDICT_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/targetDict.txt'
+TRGDICT_MD5 = 'd8c7f03ceb5fc2e5a0fa7503a4353751'
+EMB_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/emb'
+EMB_MD5 = 'bf436eb0faa1f6f9103017f8be57cdb7'
+
+UNK_IDX = 0
+
+
+def load_dict(filename):
+    d = dict()
+    with open(filename, 'r') as f:
+        for i, line in enumerate(f):
+            d[line.strip()] = i
+    return d
+
+
+def corpus_reader(data_path, words_name, props_name):
+    """
+    Read one corpus. It returns an iterator. Each element of
+    this iterator is a tuple including sentence and labels. The sentence is
+    consist of a list of word IDs. The labels include a list of label IDs.
+    :return: a iterator of data.
+    :rtype: iterator
+    """
+
+    def reader():
+        tf = tarfile.open(data_path)
+        wf = tf.extractfile(words_name)
+        pf = tf.extractfile(props_name)
+        with gzip.GzipFile(fileobj=wf) as words_file, gzip.GzipFile(
+                fileobj=pf) as props_file:
+            sentences = []
+            labels = []
+            one_seg = []
+            for word, label in itertools.izip(words_file, props_file):
+                word = word.strip()
+                label = label.strip().split()
+
+                if len(label) == 0:  # end of sentence
+                    for i in xrange(len(one_seg[0])):
+                        a_kind_lable = [x[i] for x in one_seg]
+                        labels.append(a_kind_lable)
+
+                    if len(labels) >= 1:
+                        verb_list = []
+                        for x in labels[0]:
+                            if x != '-':
+                                verb_list.append(x)
+
+                        for i, lbl in enumerate(labels[1:]):
+                            cur_tag = 'O'
+                            is_in_bracket = False
+                            lbl_seq = []
+                            verb_word = ''
+                            for l in lbl:
+                                if l == '*' and is_in_bracket == False:
+                                    lbl_seq.append('O')
+                                elif l == '*' and is_in_bracket == True:
+                                    lbl_seq.append('I-' + cur_tag)
+                                elif l == '*)':
+                                    lbl_seq.append('I-' + cur_tag)
+                                    is_in_bracket = False
+                                elif l.find('(') != -1 and l.find(')') != -1:
+                                    cur_tag = l[1:l.find('*')]
+                                    lbl_seq.append('B-' + cur_tag)
+                                    is_in_bracket = False
+                                elif l.find('(') != -1 and l.find(')') == -1:
+                                    cur_tag = l[1:l.find('*')]
+                                    lbl_seq.append('B-' + cur_tag)
+                                    is_in_bracket = True
+                                else:
+                                    raise RuntimeError('Unexpected label: %s' %
+                                                       l)
+
+                            yield sentences, verb_list[i], lbl_seq
+
+                    sentences = []
+                    labels = []
+                    one_seg = []
+                else:
+                    sentences.append(word)
+                    one_seg.append(label)
+
+        pf.close()
+        wf.close()
+        tf.close()
+
+    return reader
+
+
+def reader_creator(corpus_reader,
+                   word_dict=None,
+                   predicate_dict=None,
+                   label_dict=None):
+    def reader():
+        for sentence, predicate, labels in corpus_reader():
+
+            sen_len = len(sentence)
+
+            verb_index = labels.index('B-V')
+            mark = [0] * len(labels)
+            if verb_index > 0:
+                mark[verb_index - 1] = 1
+                ctx_n1 = sentence[verb_index - 1]
+            else:
+                ctx_n1 = 'bos'
+
+            if verb_index > 1:
+                mark[verb_index - 2] = 1
+                ctx_n2 = sentence[verb_index - 2]
+            else:
+                ctx_n2 = 'bos'
+
+            mark[verb_index] = 1
+            ctx_0 = sentence[verb_index]
+
+            if verb_index < len(labels) - 1:
+                mark[verb_index + 1] = 1
+                ctx_p1 = sentence[verb_index + 1]
+            else:
+                ctx_p1 = 'eos'
+
+            if verb_index < len(labels) - 2:
+                mark[verb_index + 2] = 1
+                ctx_p2 = sentence[verb_index + 2]
+            else:
+                ctx_p2 = 'eos'
+
+            word_idx = [word_dict.get(w, UNK_IDX) for w in sentence]
+
+            ctx_n2_idx = [word_dict.get(ctx_n2, UNK_IDX)] * sen_len
+            ctx_n1_idx = [word_dict.get(ctx_n1, UNK_IDX)] * sen_len
+            ctx_0_idx = [word_dict.get(ctx_0, UNK_IDX)] * sen_len
+            ctx_p1_idx = [word_dict.get(ctx_p1, UNK_IDX)] * sen_len
+            ctx_p2_idx = [word_dict.get(ctx_p2, UNK_IDX)] * sen_len
+
+            pred_idx = [predicate_dict.get(predicate)] * sen_len
+            label_idx = [label_dict.get(w) for w in labels]
+
+            yield word_idx, ctx_n2_idx, ctx_n1_idx, \
+              ctx_0_idx, ctx_p1_idx, ctx_p2_idx, pred_idx, mark, label_idx
+
+    return reader
+
+
+def get_dict():
+    word_dict = load_dict(download(WORDDICT_URL, 'conll05st', WORDDICT_MD5))
+    verb_dict = load_dict(download(VERBDICT_URL, 'conll05st', VERBDICT_MD5))
+    label_dict = load_dict(download(TRGDICT_URL, 'conll05st', TRGDICT_MD5))
+    return word_dict, verb_dict, label_dict
+
+
+def get_embedding():
+    return download(EMB_URL, 'conll05st', EMB_MD5)
+
+
+def test():
+    word_dict, verb_dict, label_dict = get_dict()
+    reader = corpus_reader(
+        download(DATA_URL, 'conll05st', DATA_MD5),
+        words_name='conll05st-release/test.wsj/words/test.wsj.words.gz',
+        props_name='conll05st-release/test.wsj/props/test.wsj.props.gz')
+    return reader_creator(reader, word_dict, verb_dict, label_dict)

python/paddle/v2/dataset/imdb.py

-# /usr/bin/env python
-# -*- coding:utf-8 -*-
-
 # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,6 +14,7 @@
 """
 IMDB dataset: http://ai.stanford.edu/%7Eamaas/data/sentiment/aclImdb_v1.tar.gz
 """
+
 import paddle.v2.dataset.common
 import tarfile
 import Queue
@@ -118,3 +116,8 @@ def test(word_idx):
     return reader_creator(
         re.compile("aclImdb/test/pos/.*\.txt$"),
         re.compile("aclImdb/test/neg/.*\.txt$"), word_idx, 1000)
+
+
+def word_dict():
+    return build_dict(
+        re.compile("aclImdb/((train)|(test))/((pos)|(neg))/.*\.txt$"), 150)

python/paddle/v2/dataset/imikolov.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
 """
 imikolov's simple dataset: http://www.fit.vutbr.cz/~imikolov/rnnlm/
 """
 import paddle.v2.dataset.common
 import tarfile
 
-__all__ = ['train', 'test']
+__all__ = ['train', 'test', 'build_dict']
 
 URL = 'http://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz'
 MD5 = '30177ea32e27c525793142b6bf2c8e2d'
@@ -24,7 +37,9 @@ def word_count(f, word_freq=None):
     return word_freq
 
 
-def build_dict(train_filename, test_filename):
+def build_dict():
+    train_filename = './simple-examples/data/ptb.train.txt'
+    test_filename = './simple-examples/data/ptb.valid.txt'
     with tarfile.open(
             paddle.v2.dataset.common.download(
                 paddle.v2.dataset.imikolov.URL, 'imikolov',
@@ -32,27 +47,22 @@ def build_dict(train_filename, test_filename):
         trainf = tf.extractfile(train_filename)
         testf = tf.extractfile(test_filename)
         word_freq = word_count(testf, word_count(trainf))
+        if '<unk>' in word_freq:
+            # remove <unk> for now, since we will set it as last index
+            del word_freq['<unk>']
 
         TYPO_FREQ = 50
         word_freq = filter(lambda x: x[1] > TYPO_FREQ, word_freq.items())
 
-        dictionary = sorted(word_freq, key=lambda x: (-x[1], x[0]))
-        words, _ = list(zip(*dictionary))
+        word_freq_sorted = sorted(word_freq, key=lambda x: (-x[1], x[0]))
+        words, _ = list(zip(*word_freq_sorted))
         word_idx = dict(zip(words, xrange(len(words))))
         word_idx['<unk>'] = len(words)
 
     return word_idx
 
 
-word_idx = {}
-
-
-def reader_creator(filename, n):
-    global word_idx
-    if len(word_idx) == 0:
-        word_idx = build_dict('./simple-examples/data/ptb.train.txt',
-                              './simple-examples/data/ptb.valid.txt')
-
+def reader_creator(filename, word_idx, n):
     def reader():
         with tarfile.open(
                 paddle.v2.dataset.common.download(
@@ -71,9 +81,9 @@ def reader_creator(filename, n):
     return reader
 
 
-def train(n):
-    return reader_creator('./simple-examples/data/ptb.train.txt', n)
+def train(word_idx, n):
+    return reader_creator('./simple-examples/data/ptb.train.txt', word_idx, n)
 
 
-def test(n):
-    return reader_creator('./simple-examples/data/ptb.valid.txt', n)
+def test(word_idx, n):
+    return reader_creator('./simple-examples/data/ptb.valid.txt', word_idx, n)

python/paddle/v2/dataset/mnist.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
 """
 MNIST dataset.
 """

python/paddle/v2/dataset/movielens.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import zipfile
 from common import download
 import re

python/paddle/v2/dataset/sentiment.py

+# /usr/bin/env python
+# -*- coding:utf-8 -*-
+
+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+"""
+The script fetch and preprocess movie_reviews data set
+
+that provided by NLTK
+"""
+
+import common
+import collections
+import nltk
+import numpy as np
+from itertools import chain
+from nltk.corpus import movie_reviews
+
+__all__ = ['train', 'test', 'get_word_dict']
+NUM_TRAINING_INSTANCES = 1600
+NUM_TOTAL_INSTANCES = 2000
+
+
+def download_data_if_not_yet():
+    """
+    Download the data set, if the data set is not download.
+    """
+    try:
+        # make sure that nltk can find the data
+        if common.DATA_HOME not in nltk.data.path:
+            nltk.data.path.append(common.DATA_HOME)
+        movie_reviews.categories()
+    except LookupError:
+        print "Downloading movie_reviews data set, please wait....."
+        nltk.download('movie_reviews', download_dir=common.DATA_HOME)
+        print "Download data set success....."
+        print "Path is " + nltk.data.find('corpora/movie_reviews').path
+
+
+def get_word_dict():
+    """
+    Sorted the words by the frequency of words which occur in sample
+    :return:
+        words_freq_sorted
+    """
+    words_freq_sorted = list()
+    word_freq_dict = collections.defaultdict(int)
+    download_data_if_not_yet()
+
+    for category in movie_reviews.categories():
+        for field in movie_reviews.fileids(category):
+            for words in movie_reviews.words(field):
+                word_freq_dict[words] += 1
+    words_sort_list = word_freq_dict.items()
+    words_sort_list.sort(cmp=lambda a, b: b[1] - a[1])
+    for index, word in enumerate(words_sort_list):
+        words_freq_sorted.append((word[0], index))
+    return words_freq_sorted
+
+
+def sort_files():
+    """
+    Sorted the sample for cross reading the sample
+    :return:
+        files_list
+    """
+    files_list = list()
+    neg_file_list = movie_reviews.fileids('neg')
+    pos_file_list = movie_reviews.fileids('pos')
+    files_list = list(chain.from_iterable(zip(neg_file_list, pos_file_list)))
+    return files_list
+
+
+def load_sentiment_data():
+    """
+    Load the data set
+    :return:
+        data_set
+    """
+    data_set = list()
+    download_data_if_not_yet()
+    words_ids = dict(get_word_dict())
+    for sample_file in sort_files():
+        words_list = list()
+        category = 0 if 'neg' in sample_file else 1
+        for word in movie_reviews.words(sample_file):
+            words_list.append(words_ids[word.lower()])
+        data_set.append((words_list, category))
+    return data_set
+
+
+def reader_creator(data):
+    """
+    Reader creator, generate an iterator for data set
+    :param data:
+        train data set or test data set
+    """
+    for each in data:
+        yield each[0], each[1]
+
+
+def train():
+    """
+    Default train set reader creator
+    """
+    data_set = load_sentiment_data()
+    return reader_creator(data_set[0:NUM_TRAINING_INSTANCES])
+
+
+def test():
+    """
+    Default test set reader creator
+    """
+    data_set = load_sentiment_data()
+    return reader_creator(data_set[NUM_TRAINING_INSTANCES:])

python/paddle/v2/dataset/tests/cifar_test.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import paddle.v2.dataset.cifar
 import unittest
 

python/paddle/v2/dataset/tests/common_test.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import paddle.v2.dataset.common
 import unittest
 import tempfile

python/paddle/v2/dataset/tests/imdb_test.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import paddle.v2.dataset.imdb
 import unittest
 import re

python/paddle/v2/dataset/tests/imikolov_test.py

 import paddle.v2.dataset.imikolov
 import unittest
 
+WORD_DICT = paddle.v2.dataset.imikolov.build_dict()
+
 
 class TestMikolov(unittest.TestCase):
     def check_reader(self, reader, n):
@@ -9,11 +11,15 @@ class TestMikolov(unittest.TestCase):
 
     def test_train(self):
         n = 5
-        self.check_reader(paddle.v2.dataset.imikolov.train(n), n)
+        self.check_reader(paddle.v2.dataset.imikolov.train(WORD_DICT, n), n)
 
     def test_test(self):
         n = 5
-        self.check_reader(paddle.v2.dataset.imikolov.test(n), n)
+        self.check_reader(paddle.v2.dataset.imikolov.test(WORD_DICT, n), n)
+
+    def test_total(self):
+        _, idx = zip(*WORD_DICT.items())
+        self.assertEqual(sorted(idx)[-1], len(WORD_DICT) - 1)
 
 
 if __name__ == '__main__':

python/paddle/v2/dataset/tests/mnist_test.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
 import paddle.v2.dataset.mnist
 import unittest
 

python/paddle/v2/dataset/tests/test_sentiment.py

+# /usr/bin/env python
+# -*- coding:utf-8 -*-
+
+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+import unittest
+import nltk
+import paddle.v2.dataset.sentiment as st
+from nltk.corpus import movie_reviews
+
+
+class TestSentimentMethods(unittest.TestCase):
+    def test_get_word_dict(self):
+        word_dict = st.get_word_dict()[0:10]
+        test_word_list = [(u',', 0), (u'the', 1), (u'.', 2), (u'a', 3),
+                          (u'and', 4), (u'of', 5), (u'to', 6), (u"'", 7),
+                          (u'is', 8), (u'in', 9)]
+        for idx, each in enumerate(word_dict):
+            self.assertEqual(each, test_word_list[idx])
+        self.assertTrue("/root/.cache/paddle/dataset" in nltk.data.path)
+
+    def test_sort_files(self):
+        last_label = ''
+        for sample_file in st.sort_files():
+            current_label = sample_file.split("/")[0]
+            self.assertNotEqual(current_label, last_label)
+            last_label = current_label
+
+    def test_data_set(self):
+        data_set = st.load_sentiment_data()
+        last_label = -1
+        for each in st.test():
+            self.assertNotEqual(each[1], last_label)
+            last_label = each[1]
+        self.assertEqual(len(data_set), st.NUM_TOTAL_INSTANCES)
+        self.assertEqual(len(list(st.train())), st.NUM_TRAINING_INSTANCES)
+        self.assertEqual(
+            len(list(st.test())),
+            (st.NUM_TOTAL_INSTANCES - st.NUM_TRAINING_INSTANCES))
+
+
+if __name__ == '__main__':
+    unittest.main()

python/paddle/v2/dataset/uci_housing.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+import numpy as np
+import os
+from common import download
+
+__all__ = ['train', 'test']
+
+URL = 'https://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data'
+MD5 = 'd4accdce7a25600298819f8e28e8d593'
+feature_names = [
+    'CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD', 'TAX',
+    'PTRATIO', 'B', 'LSTAT'
+]
+
+UCI_TRAIN_DATA = None
+UCI_TEST_DATA = None
+
+
+def feature_range(maximums, minimums):
+    import matplotlib
+    matplotlib.use('Agg')
+    import matplotlib.pyplot as plt
+    fig, ax = plt.subplots()
+    feature_num = len(maximums)
+    ax.bar(range(feature_num), maximums - minimums, color='r', align='center')
+    ax.set_title('feature scale')
+    plt.xticks(range(feature_num), feature_names)
+    plt.xlim([-1, feature_num])
+    fig.set_figheight(6)
+    fig.set_figwidth(10)
+    if not os.path.exists('./image'):
+        os.makedirs('./image')
+    fig.savefig('image/ranges.png', dpi=48)
+    plt.close(fig)
+
+
+def load_data(filename, feature_num=14, ratio=0.8):
+    global UCI_TRAIN_DATA, UCI_TEST_DATA
+    if UCI_TRAIN_DATA is not None and UCI_TEST_DATA is not None:
+        return
+
+    data = np.fromfile(filename, sep=' ')
+    data = data.reshape(data.shape[0] / feature_num, feature_num)
+    maximums, minimums, avgs = data.max(axis=0), data.min(axis=0), data.sum(
+        axis=0) / data.shape[0]
+    feature_range(maximums[:-1], minimums[:-1])
+    for i in xrange(feature_num - 1):
+        data[:, i] = (data[:, i] - avgs[i]) / (maximums[i] - minimums[i])
+    offset = int(data.shape[0] * ratio)
+    UCI_TRAIN_DATA = data[:offset]
+    UCI_TEST_DATA = data[offset:]
+
+
+def train():
+    global UCI_TRAIN_DATA
+    load_data(download(URL, 'uci_housing', MD5))
+
+    def reader():
+        for d in UCI_TRAIN_DATA:
+            yield d[:-1], d[-1:]
+
+    return reader
+
+
+def test():
+    global UCI_TEST_DATA
+    load_data(download(URL, 'uci_housing', MD5))
+
+    def reader():
+        for d in UCI_TEST_DATA:
+            yield d[:-1], d[-1:]
+
+    return reader

python/paddle/v2/dataset/wmt14.py

+# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+"""
+wmt14 dataset
+"""
+import paddle.v2.dataset.common
+import tarfile
+import os.path
+import itertools
+
+__all__ = ['train', 'test', 'build_dict']
+
+URL_DEV_TEST = 'http://www-lium.univ-lemans.fr/~schwenk/cslm_joint_paper/data/dev+test.tgz'
+MD5_DEV_TEST = '7d7897317ddd8ba0ae5c5fa7248d3ff5'
+URL_TRAIN = 'http://localhost:8000/train.tgz'
+MD5_TRAIN = '72de99da2830ea5a3a2c4eb36092bbc7'
+
+
+def word_count(f, word_freq=None):
+    add = paddle.v2.dataset.common.dict_add
+    if word_freq == None:
+        word_freq = {}
+
+    for l in f:
+        for w in l.strip().split():
+            add(word_freq, w)
+        add(word_freq, '<s>')
+        add(word_freq, '<e>')
+
+    return word_freq
+
+
+def get_word_dix(word_freq):
+    TYPO_FREQ = 50
+    word_freq = filter(lambda x: x[1] > TYPO_FREQ, word_freq.items())
+    word_freq_sorted = sorted(word_freq, key=lambda x: (-x[1], x[0]))
+    words, _ = list(zip(*word_freq_sorted))
+    word_idx = dict(zip(words, xrange(len(words))))
+    word_idx['<unk>'] = len(words)
+    return word_idx
+
+
+def get_word_freq(train, dev):
+    word_freq = word_count(train, word_count(dev))
+    if '<unk>' in word_freq:
+        # remove <unk> for now, since we will set it as last index
+        del word_freq['<unk>']
+    return word_freq
+
+
+def build_dict():
+    base_dir = './wmt14-data'
+    train_en_filename = base_dir + '/train/train.en'
+    train_fr_filename = base_dir + '/train/train.fr'
+    dev_en_filename = base_dir + '/dev/ntst1213.en'
+    dev_fr_filename = base_dir + '/dev/ntst1213.fr'
+
+    if not os.path.exists(train_en_filename) or not os.path.exists(
+            train_fr_filename):
+        with tarfile.open(
+                paddle.v2.dataset.common.download(URL_TRAIN, 'wmt14',
+                                                  MD5_TRAIN)) as tf:
+            tf.extractall(base_dir)
+
+    if not os.path.exists(dev_en_filename) or not os.path.exists(
+            dev_fr_filename):
+        with tarfile.open(
+                paddle.v2.dataset.common.download(URL_DEV_TEST, 'wmt14',
+                                                  MD5_DEV_TEST)) as tf:
+            tf.extractall(base_dir)
+
+    f_en = open(train_en_filename)
+    f_fr = open(train_fr_filename)
+    f_en_dev = open(dev_en_filename)
+    f_fr_dev = open(dev_fr_filename)
+
+    word_freq_en = get_word_freq(f_en, f_en_dev)
+    word_freq_fr = get_word_freq(f_fr, f_fr_dev)
+
+    f_en.close()
+    f_fr.close()
+    f_en_dev.close()
+    f_fr_dev.close()
+
+    return get_word_dix(word_freq_en), get_word_dix(word_freq_fr)
+
+
+def reader_creator(directory, path_en, path_fr, URL, MD5, dict_en, dict_fr):
+    def reader():
+        if not os.path.exists(path_en) or not os.path.exists(path_fr):
+            with tarfile.open(
+                    paddle.v2.dataset.common.download(URL, 'wmt14', MD5)) as tf:
+                tf.extractall(directory)
+
+        f_en = open(path_en)
+        f_fr = open(path_fr)
+        UNK_en = dict_en['<unk>']
+        UNK_fr = dict_fr['<unk>']
+
+        for en, fr in itertools.izip(f_en, f_fr):
+            src_ids = [dict_en.get(w, UNK_en) for w in en.strip().split()]
+            tar_ids = [
+                dict_fr.get(w, UNK_fr)
+                for w in ['<s>'] + fr.strip().split() + ['<e>']
+            ]
+
+            # remove sequence whose length > 80 in training mode
+            if len(src_ids) == 0 or len(tar_ids) <= 1 or len(
+                    src_ids) > 80 or len(tar_ids) > 80:
+                continue
+
+            yield src_ids, tar_ids[:-1], tar_ids[1:]
+
+        f_en.close()
+        f_fr.close()
+
+    return reader
+
+
+def train(dict_en, dict_fr):
+    directory = './wmt14-data'
+    return reader_creator(directory, directory + '/train/train.en',
+                          directory + '/train/train.fr', URL_TRAIN, MD5_TRAIN,
+                          dict_en, dict_fr)
+
+
+def test(dict_en, dict_fr):
+    directory = './wmt14-data'
+    return reader_creator(directory, directory + '/dev/ntst1213.en',
+                          directory + '/dev/ntst1213.fr', URL_DEV_TEST,
+                          MD5_DEV_TEST, dict_en, dict_fr)

python/paddle/v2/event.py

@@ -34,8 +34,9 @@ class WithMetric(object):
 
 
 class TestResult(WithMetric):
-    def __init__(self, evaluator):
+    def __init__(self, evaluator, cost):
         super(TestResult, self).__init__(evaluator)
+        self.cost = cost
 
 
 class BeginPass(object):

python/paddle/v2/inferencer.py → python/paddle/v2/inference.py

@@ -5,7 +5,7 @@ from data_feeder import DataFeeder
 import itertools
 import numpy
 
-__all__ = ['Inference', 'infer']
+__all__ = ['infer']
 
 
 class Inference(object):

python/paddle/v2/layer.py

@@ -12,72 +12,42 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-Before this new package paddle.v2.layer, users would need to use functions
-in paddle.trainer_config_helpers.layers to configure networks.
-
-The Old Way:
-=========
-This old way requires that the creation of a network be defined in a Python
-function, say network_config, and that this Python function being passed to
-paddle.trainer_config_helpers.parse_network_config for the creation of
-protobuf message description of this network.
-
-```python
-def network_config():
-  img = paddle.trainer_config_helpers.data_layer(name="pixel", size=784)
-  inference = paddle.trainer_config_helpers.fc_layer(
-    input=img,
-    size=10,
-    act=paddle.trainer_config_helpers.SoftmaxActivation())
-  cost = paddle.trainer_config_helpers.classification_cost(
-    input=inference,
-    label=paddle.trainer_config_helpers.data_layer(name="label", size=10))
-
-proto_desc = parse_network_config(network_config)
-```
-
-When parse_network_config executes network_config, those layer definition
-functions like data_layer and fc_layer would change some Python global variables,
-so that after the execution, parse_network_config could collect information from
-these global variables and generates the protobuf message.
-
-
-
-The New Way:
-=========
-In this PR, we define a function in paddle.v2.layer which creates a Python
-class for each layer creation function in paddle.trainer_config_helpers.layers.
-Users can use create a network as follows:
-
-```python
-img = paddle.v2.layer.data(name="pixel", size=784)
-inference = paddle.v2.layer.fc(input=img, size=10, act=paddle.v2.layer.Softmax())
-cost = paddle.v2.layer.classification(
-  input=inference,
-  label=paddle.v2.layer.data(name="label", size=10))
-
-parameters = paddle.v2.parameters.create(cost)
-```
-
-This new way doesn't require those invocations to layer definition functions
-to be in a Python function but could be anywhere.
-
-Also, the creation of a protobuf message is hidden in the invocation of
-paddle.v2.parameters.create, no longer exposed to users.
+`paddle.v2.layer` is a part of model config packages in paddle.v2. In API v2,
+we want to make Paddle a plain Python package. The model config package defined
+the way how to configure a neural network topology in Paddle Python code.
+
+The primary usage shows below.
+
+..  code-block:: python
+
+    import paddle.v2 as paddle
+
+    img = paddle.layer.data(name='img', type=paddle.data_type.dense_vector(784))
+    hidden = paddle.layer.fc(input=img, size=200)
+    prediction = paddle.layer.fc(input=hidden, size=10,
+                                 act=paddle.activation.Softmax())
+
+    # use prediction instance where needed.
+    parameters = paddle.v2.parameters.create(cost)
 """
+
+import collections
+import inspect
 from config_base import Layer, __convert_to_v2__
 import paddle.trainer_config_helpers as conf_helps
 from paddle.trainer_config_helpers.config_parser_utils import \
     parse_network_config as __parse__
-
-from paddle.trainer_config_helpers.default_decorators import wrap_name_default
 from paddle.trainer_config_helpers.default_decorators import wrap_act_default
 from paddle.trainer_config_helpers.default_decorators import \
     wrap_bias_attr_default
+from paddle.trainer_config_helpers.default_decorators import wrap_name_default
 from paddle.trainer_config_helpers.layers import layer_support
+from paddle.trainer.config_parser import \
+    RecurrentLayerGroupWithoutOutLinksBegin, RecurrentLayerGroupSetOutLink, \
+    RecurrentLayerGroupEnd, model_type
 
-import data_type
 import activation
+import data_type
 
 __all__ = ['parse_network', 'data']
 
@@ -130,6 +100,137 @@ class DataLayerV2(Layer):
         return getattr(conf_helps, self.__method_name__)(name=self.name, **args)
 
 
+class WithExtraParent(Layer):
+    def extra_parent(self):
+        return self.__extra_parent__
+
+    def __init__(self, name=None, parent_layers=None):
+        self.__extra_parent__ = []
+        super(WithExtraParent, self).__init__(
+            name=name, parent_layers=parent_layers)
+
+    def append_extra_parent(self, parent):
+        self.__extra_parent__.append(parent)
+
+    def to_proto(self, context):
+        """
+        function to set proto attribute
+        """
+        kwargs = dict()
+        for p in self.__extra_parent__:
+            p.to_proto(context=context)
+
+        for layer_name in self.__parent_layers__:
+            if not isinstance(self.__parent_layers__[layer_name],
+                              collections.Sequence):
+                v1_layer = self.__parent_layers__[layer_name].to_proto(
+                    context=context)
+            else:
+                v1_layer = map(lambda x: x.to_proto(context=context),
+                               self.__parent_layers__[layer_name])
+            kwargs[layer_name] = v1_layer
+
+        if self.context_name() is None:
+            return self.to_proto_impl(context=context, **kwargs)
+        elif self.context_name() not in context:
+            context[self.context_name()] = self.to_proto_impl(
+                context=context, **kwargs)
+
+        if self.use_context_name():
+            return context[self.context_name()]
+        else:
+            return context[self.name]
+
+
+class MemoryV2(WithExtraParent):
+    def __init__(self, name, **kwargs):
+        self.name = name
+        super(MemoryV2, self).__init__(name=name, parent_layers=dict())
+        self.__kwargs__ = kwargs
+        self.__boot_layer_name__ = None
+        if 'boot_layer' in kwargs:
+            begin_of_current_rnn = []
+            # TODO(yuyang18): Fix inspect, it could be wrong when user invoke a
+            # function inside step.
+            st = inspect.stack()
+            for i in xrange(len(st)):
+                locs = inspect.stack()[i][0].f_locals
+                keys = locs.keys()
+                for key in keys:
+                    val = locs[key]
+                    if isinstance(val, RecurrentLayerInput):
+                        begin_of_current_rnn.append(val)
+                    elif isinstance(val, collections.Sequence):
+                        for v in val:
+                            if isinstance(v, RecurrentLayerInput):
+                                begin_of_current_rnn.append(v)
+
+                if begin_of_current_rnn:
+                    break
+            assert begin_of_current_rnn is not None
+            for extra in begin_of_current_rnn:
+                self.append_extra_parent(extra)
+                assert isinstance(extra, WithExtraParent)
+                extra.append_extra_parent(kwargs['boot_layer'])
+                self.__boot_layer_name__ = kwargs['boot_layer'].name
+
+    def to_proto_impl(self, context, **kwargs):
+        args = dict()
+        for each in kwargs:
+            args[each] = kwargs[each]
+        for each in self.__kwargs__:
+            args[each] = self.__kwargs__[each]
+
+        if self.__boot_layer_name__ is not None:
+            args['boot_layer'] = context[self.__boot_layer_name__]
+
+        size = args.get('size', None)
+        if size is not None:
+            if callable(size):
+                real_size = size()
+            else:
+                real_size = size
+            args['size'] = real_size
+        return conf_helps.memory(name=self.name, **args)
+
+    def context_name(self):
+        return self.name + "#memory"
+
+    def use_context_name(self):
+        """
+        memory layer will have the same name with some layer
+        :return:
+        """
+        return True
+
+
+class LayerOutputV2(Layer):
+    """
+    LayerOutputV2 is used to store the result of LayerOutput in v1 api.
+    It will not store it's parents because layer_output has been parsed already.
+    """
+
+    def __init__(self, layer_output):
+        assert isinstance(layer_output, conf_helps.LayerOutput)
+        self.layer_output = layer_output
+        super(LayerOutputV2, self).__init__(
+            name=layer_output.name, parent_layers=dict())
+
+    def to_proto_impl(self):
+        return self.layer_output
+
+
+class StaticInputV2(object):
+    def __init__(self, input, is_seq=False, size=None):
+        assert isinstance(input, LayerV2)
+        self.name = input.name
+        self.input = input
+        self.is_seq = is_seq
+        self.size = size
+        # TODO(add size check)
+        # assert input.size is not None or size is not None
+
+
 class MixedLayerV2(Layer):
     """
     This class is use to support `with` grammar. If not, the following code
@@ -161,7 +262,6 @@ class MixedLayerV2(Layer):
         other_kwargs['act'] = act
         other_kwargs['bias_attr'] = bias_attr
         other_kwargs['layer_attr'] = layer_attr
-
         parent_layers = {"input": self.__inputs__}
         super(MixedLayerV2, self).__init__(name, parent_layers)
         self.__other_kwargs__ = other_kwargs
@@ -171,7 +271,7 @@ class MixedLayerV2(Layer):
             self.__inputs__.append(other)
             return self
         else:
-            raise MixedLayerTypeV2.AddToSealedMixedLayerExceptionV2()
+            raise MixedLayerV2.AddToSealedMixedLayerExceptionV2()
 
     def __enter__(self):
         assert len(self.__inputs__) == 0
@@ -186,6 +286,13 @@ class MixedLayerV2(Layer):
             args[each] = kwargs[each]
         for each in self.__other_kwargs__:
             args[each] = self.__other_kwargs__[each]
+        size = args.get('size', None)
+        if size is not None:
+            if callable(size):
+                real_size = size()
+            else:
+                real_size = size
+            args['size'] = real_size
         return getattr(conf_helps, self.__method_name__)(**args)
 
 
@@ -202,14 +309,51 @@ def mixed(size=0,
     return MixedLayerV2(size, input, name, act, bias_attr, layer_attr)
 
 
+class RecurrentLayerInput(WithExtraParent):
+    def __init__(self, recurrent_name, index, parent_layers):
+        assert len(parent_layers) == 1
+        self.__parents__ = parent_layers.values()[0]
+        super(RecurrentLayerInput, self).__init__(
+            name=self.__parents__[index].name, parent_layers=parent_layers)
+        self.__recurrent_name__ = recurrent_name
+
+    def context_name(self):
+        return self.__recurrent_name__ + ".begin"
+
+    def to_proto_impl(self, context, **kwargs):
+        model_type('recurrent_nn')
+        RecurrentLayerGroupWithoutOutLinksBegin(
+            name=self.__recurrent_name__,
+            in_links=map(lambda x: x.name, self.__parents__))
+        return self
+
+
+class RecurrentLayerOutput(Layer):
+    def __init__(self, recurrent_name, index, parent_layers):
+        assert len(parent_layers) == 1
+        self.__parents__ = parent_layers.values()[0]
+        super(RecurrentLayerOutput, self).__init__(
+            name=self.__parents__[index].name, parent_layers=parent_layers)
+        self.__recurrent_name__ = recurrent_name
+
+    def context_name(self):
+        return self.__recurrent_name__ + ".end"
+
+    def to_proto_impl(self, **kwargs):
+        for l in self.__parents__:
+            RecurrentLayerGroupSetOutLink(l.name)
+        RecurrentLayerGroupEnd(name=self.__recurrent_name__)
+
+
 LayerV2 = Layer
 data = DataLayerV2
 AggregateLevel = conf_helps.layers.AggregateLevel
 ExpandLevel = conf_helps.layers.ExpandLevel
+memory = MemoryV2
 
 
 def __layer_name_mapping__(inname):
-    if inname in ['data_layer', 'memory', 'mixed_layer']:
+    if inname in ['data_layer', 'memory', 'mixed_layer', 'recurrent_group']:
         # Do Not handle these layers
         return
     elif inname == 'maxid_layer':
@@ -231,8 +375,10 @@ def __layer_name_mapping__(inname):
 def __layer_name_mapping_parent_names__(inname):
     all_args = getattr(conf_helps, inname).argspec.args
     return filter(
-        lambda x: x in ['input1', 'input2','label', 'input', 'a', 'b', 'expand_as',
-                        'weights', 'vectors', 'weight', 'score', 'left', 'right'],
+        lambda x: x in ['input1', 'input2', 'label', 'input', 'a', 'b',
+                        'expand_as',
+                        'weights', 'vectors', 'weight', 'score', 'left',
+                        'right', 'output_mem'],
         all_args)
 
 
@@ -267,3 +413,54 @@ operator_list = [
 for op in operator_list:
     globals()[op[0]] = __convert_to_v2__(
         op[0], parent_names=op[1], is_default_name=False)
+
+
+@wrap_name_default()
+def recurrent_group(step, input, name=None):
+    if not isinstance(input, collections.Sequence):
+        input = [input]
+
+    non_static_inputs = filter(lambda x: not isinstance(x, StaticInputV2),
+                               input)
+    actual_input = [
+        RecurrentLayerInput(
+            recurrent_name=name,
+            index=i,
+            parent_layers={'recurrent_inputs': non_static_inputs})
+        for i in xrange(len(non_static_inputs))
+    ]
+
+    def __real_step__(*args):
+        rnn_input = list(args)
+        static_inputs = filter(lambda x: isinstance(x, StaticInputV2), input)
+        for static_input in static_inputs:
+            mem_name = "__%s_memory__" % static_input.input.name
+            mem = memory(
+                name=mem_name,
+                is_seq=static_input.is_seq,
+                size=static_input.input.calculate_size,
+                boot_layer=static_input.input)
+            with mixed(
+                    name=mem_name,
+                    size=static_input.input.calculate_size,
+                    act=activation.Identity()) as mix:
+                mix += identity_projection(input=mem)
+            rnn_input.insert(input.index(static_input), mix)
+        return step(*rnn_input)
+
+    actual_output = __real_step__(*actual_input)
+
+    if not isinstance(actual_output, collections.Sequence):
+        actual_output = [actual_output]
+
+    retv = [
+        RecurrentLayerOutput(
+            recurrent_name=name,
+            index=i,
+            parent_layers={'recurrent_outputs': actual_output})
+        for i in xrange(len(actual_output))
+    ]
+    if len(retv) == 1:
+        return retv[0]
+    else:
+        return retv

python/paddle/v2/tests/CMakeLists.txt

+add_test(NAME test_v2_api
+        COMMAND bash ${PROJ_ROOT}/python/paddle/v2/tests/run_tests.sh ${PYTHON_EXECUTABLE})
+
 add_test(NAME test_v2_layer
         COMMAND ${PROJ_ROOT}/paddle/.set_python_path.sh -d ${PROJ_ROOT}/python/
         ${PYTHON_EXECUTABLE} ${PROJ_ROOT}/python/paddle/v2/tests/test_layer.py
         WORKING_DIRECTORY ${PROJ_ROOT}/python/paddle)
 
-add_test(NAME test_v2_api
-        COMMAND bash ${PROJ_ROOT}/python/paddle/v2/tests/run_tests.sh ${PYTHON_EXECUTABLE})
+add_test(NAME test_v2_rnn_layer
+        COMMAND ${PROJ_ROOT}/paddle/.set_python_path.sh -d ${PROJ_ROOT}/python/
+        ${PYTHON_EXECUTABLE} ${PROJ_ROOT}/python/paddle/v2/tests/test_rnn_layer.py)
 
-add_test(NAME topology_test
+add_test(NAME test_topology
         COMMAND ${PROJ_ROOT}/paddle/.set_python_path.sh -d ${PROJ_ROOT}/python/
         ${PYTHON_EXECUTABLE} ${PROJ_ROOT}/python/paddle/v2/tests/test_topology.py
         WORKING_DIRECTORY ${PROJ_ROOT}/python/paddle)

python/paddle/v2/tests/test_data_feeder.py

@@ -110,14 +110,14 @@ class DataFeederTest(unittest.TestCase):
             self.assertAlmostEqual(value.all(), w[i].all())
 
     def test_integer(self):
-        dim = 100
+        value_range = 100
         batch_size = 32
         index = []
         for i in xrange(batch_size):
             each_sample = []
-            each_sample.append(np.random.randint(dim))
+            each_sample.append(np.random.randint(value_range))
             index.append(each_sample)
-        feeder = DataFeeder([('input', data_type.integer_value(dim))],
+        feeder = DataFeeder([('input', data_type.integer_value(value_range))],
                             {'input': 0})
         arg = feeder(index)
         output = arg.getSlotIds(0).copyToNumpyArray()
@@ -125,7 +125,7 @@ class DataFeederTest(unittest.TestCase):
         self.assertEqual(output.all(), index.flatten().all())
 
     def test_integer_sequence(self):
-        dim = 10000
+        value_range = 10000
         batch_size = 32
         start = [0]
         data = []
@@ -133,11 +133,12 @@ class DataFeederTest(unittest.TestCase):
             each_sample = []
             each_sample.append(
                 self.sparse_binary_reader(
-                    dim, 30, non_empty=True))
+                    value_range, 30, non_empty=True))
             data.append(each_sample)
             start.append(len(each_sample[0]) + start[-1])
-        feeder = DataFeeder([('input', data_type.integer_value_sequence(dim))],
-                            {'input': 0})
+        feeder = DataFeeder(
+            [('input', data_type.integer_value_sequence(value_range))],
+            {'input': 0})
         arg = feeder(data)
         output_data = arg.getSlotIds(0).copyToNumpyArray()
         output_start = arg.getSlotSequenceStartPositions(0).copyToNumpyArray()

python/paddle/v2/tests/test_rnn_layer.py

+# Copyright PaddlePaddle contributors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+import difflib
+import unittest
+
+import paddle.trainer_config_helpers as conf_helps
+import paddle.v2.activation as activation
+import paddle.v2.data_type as data_type
+import paddle.v2.layer as layer
+from paddle.trainer_config_helpers.config_parser_utils import \
+    parse_network_config as parse_network
+
+
+class RNNTest(unittest.TestCase):
+    def test_simple_rnn(self):
+        dict_dim = 10
+        word_dim = 8
+        hidden_dim = 8
+
+        def parse_old_rnn():
+            def step(y):
+                mem = conf_helps.memory(name="rnn_state", size=hidden_dim)
+                out = conf_helps.fc_layer(
+                    input=[y, mem],
+                    size=hidden_dim,
+                    act=activation.Tanh(),
+                    bias_attr=True,
+                    name="rnn_state")
+                return out
+
+            def test():
+                data = conf_helps.data_layer(name="word", size=dict_dim)
+                embd = conf_helps.embedding_layer(input=data, size=word_dim)
+                conf_helps.recurrent_group(name="rnn", step=step, input=embd)
+
+            return str(parse_network(test))
+
+        def parse_new_rnn():
+            def new_step(y):
+                mem = layer.memory(name="rnn_state", size=hidden_dim)
+                out = layer.fc(input=[y, mem],
+                               size=hidden_dim,
+                               act=activation.Tanh(),
+                               bias_attr=True,
+                               name="rnn_state")
+                return out
+
+            data = layer.data(
+                name="word", type=data_type.integer_value(dict_dim))
+            embd = layer.embedding(input=data, size=word_dim)
+            rnn_layer = layer.recurrent_group(
+                name="rnn", step=new_step, input=embd)
+            return str(layer.parse_network(rnn_layer))
+
+        diff = difflib.unified_diff(parse_old_rnn().splitlines(1),
+                                    parse_new_rnn().splitlines(1))
+        print ''.join(diff)
+
+    def test_sequence_rnn_multi_input(self):
+        dict_dim = 10
+        word_dim = 8
+        hidden_dim = 8
+        label_dim = 3
+
+        def parse_old_rnn():
+            def test():
+                data = conf_helps.data_layer(name="word", size=dict_dim)
+                label = conf_helps.data_layer(name="label", size=label_dim)
+                emb = conf_helps.embedding_layer(input=data, size=word_dim)
+                boot_layer = conf_helps.data_layer(name="boot", size=10)
+                boot_layer = conf_helps.fc_layer(
+                    name='boot_fc', input=boot_layer, size=10)
+
+                def step(y, wid):
+                    z = conf_helps.embedding_layer(input=wid, size=word_dim)
+                    mem = conf_helps.memory(
+                        name="rnn_state",
+                        size=hidden_dim,
+                        boot_layer=boot_layer)
+                    out = conf_helps.fc_layer(
+                        input=[y, z, mem],
+                        size=hidden_dim,
+                        act=conf_helps.TanhActivation(),
+                        bias_attr=True,
+                        name="rnn_state")
+                    return out
+
+                out = conf_helps.recurrent_group(
+                    name="rnn", step=step, input=[emb, data])
+
+                rep = conf_helps.last_seq(input=out)
+                prob = conf_helps.fc_layer(
+                    size=label_dim,
+                    input=rep,
+                    act=conf_helps.SoftmaxActivation(),
+                    bias_attr=True)
+
+                conf_helps.outputs(
+                    conf_helps.classification_cost(
+                        input=prob, label=label))
+
+            return str(parse_network(test))
+
+        def parse_new_rnn():
+            data = layer.data(
+                name="word", type=data_type.dense_vector(dict_dim))
+            label = layer.data(
+                name="label", type=data_type.dense_vector(label_dim))
+            emb = layer.embedding(input=data, size=word_dim)
+            boot_layer = layer.data(
+                name="boot", type=data_type.dense_vector(10))
+            boot_layer = layer.fc(name='boot_fc', input=boot_layer, size=10)
+
+            def step(y, wid):
+                z = layer.embedding(input=wid, size=word_dim)
+                mem = layer.memory(
+                    name="rnn_state", size=hidden_dim, boot_layer=boot_layer)
+                out = layer.fc(input=[y, z, mem],
+                               size=hidden_dim,
+                               act=activation.Tanh(),
+                               bias_attr=True,
+                               name="rnn_state")
+                return out
+
+            out = layer.recurrent_group(
+                name="rnn", step=step, input=[emb, data])
+
+            rep = layer.last_seq(input=out)
+            prob = layer.fc(size=label_dim,
+                            input=rep,
+                            act=activation.Softmax(),
+                            bias_attr=True)
+
+            cost = layer.classification_cost(input=prob, label=label)
+
+            return str(layer.parse_network(cost))
+
+        diff = difflib.unified_diff(parse_old_rnn().splitlines(1),
+                                    parse_new_rnn().splitlines(1))
+        print ''.join(diff)
+
+
+if __name__ == '__main__':
+    unittest.main()

python/paddle/v2/topology.py

@@ -17,6 +17,7 @@ import collections
 from paddle.proto.ModelConfig_pb2 import ModelConfig
 
 import layer as v2_layer
+from layer import WithExtraParent
 
 __all__ = ['Topology']
 
@@ -40,7 +41,10 @@ def __bfs_travel__(callback, *layers):
         __break__ = callback(each_layer)
         if __break__:
             return
-        __bfs_travel__(callback, *each_layer.__parent_layers__.values())
+        __layers__ = each_layer.__parent_layers__.values()
+        if isinstance(each_layer, WithExtraParent):
+            __layers__ = __layers__ + each_layer.extra_parent()
+        __bfs_travel__(callback, *__layers__)
 
 
 class Topology(object):

python/paddle/v2/trainer.py

@@ -8,7 +8,7 @@ from . import event as v2_event
 from . import optimizer as v2_optimizer
 from . import parameters as v2_parameters
 
-__all__ = ['ITrainer', 'SGD']
+__all__ = ['SGD']
 
 
 def default_event_handler(event):
@@ -22,26 +22,7 @@ def default_event_handler(event):
     pass
 
 
-class ITrainer(object):
-    """
-    The interface of Trainer. The only exposed method is `train`.
-    """
-
-    def train(self, reader, topology, parameters, event_handler=None):
-        """
-        train method.
-
-        :param reader:
-        :param topology:
-        :param parameters:
-        :param event_handler:
-        :return:
-        """
-
-        raise NotImplementedError()
-
-
-class SGD(ITrainer):
+class SGD():
     def __init__(self, cost, parameters, update_equation):
         """
         Simple SGD Trainer.
@@ -108,9 +89,6 @@ class SGD(ITrainer):
             pass_evaluator.start()
             updater.startPass()
             for batch_id, data_batch in enumerate(reader()):
-                pass_type = updater.startBatch(len(data_batch))
-                self.__gradient_machine__.forwardBackward(
-                    feeder(data_batch), out_args, pass_type)
                 batch_evaluator.start()
                 event_handler(
                     v2_event.BeginIteration(
@@ -120,12 +98,11 @@ class SGD(ITrainer):
                     feeder(data_batch), out_args, pass_type)
                 self.__gradient_machine__.eval(pass_evaluator)
                 self.__gradient_machine__.eval(batch_evaluator)
-                for each_param in self.__gradient_machine__.getParameters():
+                for each_param in self.__gradient_machine__.getNonStaticParameters(
+                ):
                     updater.update(each_param)
-                # Get cost. We use numpy to calculate total cost for this batch.
-                cost_vec = out_args.getSlotValue(0)
-                cost_vec = cost_vec.copyToNumpyMat()
-                cost = cost_vec.sum() / len(data_batch)
+                cost_sum = out_args.sumCosts()
+                cost = cost_sum / len(data_batch)
                 updater.finishBatch(cost)
                 batch_evaluator.finish()
                 event_handler(
@@ -154,13 +131,18 @@ class SGD(ITrainer):
         evaluator = self.__gradient_machine__.makeEvaluator()
         out_args = api.Arguments.createArguments(0)
         evaluator.start()
+        total_cost = 0
+        num_samples = 0.0
         for data_batch in reader():
+            num_samples += len(data_batch)
             self.__gradient_machine__.forward(
                 feeder(data_batch), out_args, api.PASS_TEST)
+            total_cost += out_args.sumCosts()
             self.__gradient_machine__.eval(evaluator)
 
         evaluator.finish()
-        return v2_event.TestResult(evaluator=evaluator)
+        return v2_event.TestResult(
+            evaluator=evaluator, cost=total_cost / num_samples)
 
 
 def __check_train_args__(reader, event_handler, **kwargs):