Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into srl_api_v2

demo/mnist/api_train_v2.py

-import numpy
 import paddle.v2 as paddle
 
 import mnist_util
@@ -40,17 +39,14 @@ def main():
 
     trainer = paddle.trainer.SGD(update_equation=adam_optimizer)
 
-    trainer.train(train_data_reader=train_reader,
-                  topology=cost,
-                  parameters=parameters,
-                  event_handler=event_handler,
-                  batch_size=32,  # batch size should be refactor in Data reader
-                  data_types=[  # data_types will be removed, It should be in
-                      # network topology
-                      ('pixel', images.type),
-                      ('label', label.type)],
-                  reader_dict={'pixel':0, 'label':1}
-                  )
+    trainer.train(
+        train_data_reader=train_reader,
+        cost=cost,
+        parameters=parameters,
+        event_handler=event_handler,
+        batch_size=32,  # batch size should be refactor in Data reader
+        reader_dict={images.name: 0,
+                     label.name: 1})
 
 
 if __name__ == '__main__':

doc/api/trainer_config_helpers/layers.rst

@@ -139,24 +139,12 @@ lstmemory
     :members: lstmemory
     :noindex:
 
-lstm_step_layer
----------------
-..  automodule:: paddle.trainer_config_helpers.layers
-    :members: lstm_step_layer
-    :noindex:
-
 grumemory
 ---------
 ..  automodule:: paddle.trainer_config_helpers.layers
     :members: grumemory
     :noindex:
 
-gru_step_layer
----------------
-..  automodule:: paddle.trainer_config_helpers.layers
-    :members: gru_step_layer
-    :noindex:
-
 Recurrent Layer Group
 =====================
 
@@ -172,6 +160,18 @@ recurrent_group
     :members: recurrent_group
     :noindex:
     
+lstm_step_layer
+---------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: lstm_step_layer
+    :noindex:
+
+gru_step_layer
+---------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: gru_step_layer
+    :noindex:
+
 beam_search
 ------------
 ..  automodule:: paddle.trainer_config_helpers.layers
@@ -308,6 +308,12 @@ repeat_layer
     :members: repeat_layer
     :noindex:
 
+rotate_layer
+------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: rotate_layer
+    :noindex:
+
 seq_reshape_layer
 -----------------
 ..  automodule:: paddle.trainer_config_helpers.layers
@@ -462,6 +468,12 @@ ctc_layer
     :members: ctc_layer
     :noindex:
 
+warp_ctc_layer
+--------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: warp_ctc_layer
+    :noindex:
+
 nce_layer
 -----------
 ..  automodule:: paddle.trainer_config_helpers.layers

python/paddle/trainer_config_helpers/layers.py

@@ -112,6 +112,7 @@ __all__ = [
     'priorbox_layer',
     'spp_layer',
     'pad_layer',
+    'eos_layer',
     'layer_support',
 ]
 
@@ -1289,6 +1290,12 @@ def last_seq(input,
     """
     Get Last Timestamp Activation of a sequence.
 
+    The simple usage is:
+
+    .. code-block:: python
+
+       seq = last_seq(input=layer)
+
     :param agg_level: Aggregated level
     :param name: Layer name.
     :type name: basestring
@@ -1327,6 +1334,12 @@ def first_seq(input,
     """
     Get First Timestamp Activation of a sequence.
 
+    The simple usage is:
+
+    .. code-block:: python
+
+       seq = first_seq(input=layer)
+
     :param agg_level: aggregation level
     :param name: Layer name.
     :type name: basestring
@@ -1427,7 +1440,7 @@ def repeat_layer(input, num_repeats, name=None, layer_attr=None):
 
     .. code-block:: python
 
-       expand = repeat_layer(layer, 4)
+       expand = repeat_layer(input=layer, num_repeats=4)
 
     :param input: Input layer
     :type input: LayerOutput
@@ -1799,6 +1812,12 @@ def cos_sim(a, b, scale=1, size=1, name=None, layer_attr=None):
     Note that the above computation is for one sample. Multiple samples are
     processed in one batch.
 
+    The example usage is:
+
+    .. code-block:: python
+
+       cos = cos_sim(a=layer1, b=layer2, size=3)
+
     :param name: layer name
     :type name: basestring
     :param a: input layer a
@@ -1960,6 +1979,16 @@ def img_conv_layer(input,
     pieces. First 256/4 = 64 channels will process by first 32 filters. The
     rest channels will be processed by rest group of filters.
 
+    The example usage is:
+
+    ..  code-block:: python
+
+        conv = img_conv_layer(input=data, filter_size=1, filter_size_y=1,
+                              num_channels=8,
+                              num_filters=16, stride=1,
+                              bias_attr=False,
+                              act=ReluActivation())
+
     :param name: Layer name.
     :type name: basestring
     :param input: Layer Input.
@@ -2099,6 +2128,34 @@ def img_pool_layer(input,
 
     .. _pooling: http://ufldl.stanford.edu/tutorial/supervised/Pooling/
 
+    - ceil_mode=True:
+
+    ..  math::
+
+        w = 1 + int(ceil(input\_width + 2 * padding - pool\_size) / float(stride))
+        h = 1 + int(ceil(input\_height + 2 * padding\_y - pool\_size\_y) / float(stride\_y))
+
+    - ceil_mode=False:
+
+    ..  math::
+
+        w = 1 + int(floor(input\_width + 2 * padding - pool\_size) / float(stride))
+        h = 1 + int(floor(input\_height + 2 * padding\_y - pool\_size\_y) / float(stride\_y))
+
+    The example usage is:
+
+    ..  code-block:: python
+
+        maxpool = img_pool_layer(input=conv,
+                                 pool_size=3,
+                                 pool_size_y=5,
+                                 num_channels=8,
+                                 stride=1,
+                                 stride_y=2,
+                                 padding=1,
+                                 padding_y=2,
+                                 pool_type=MaxPooling())
+
     :param padding: pooling padding width.
     :type padding: int
     :param padding_y: pooling padding height. It's equal to padding by default.
@@ -2125,19 +2182,6 @@ def img_pool_layer(input,
     :param ceil_mode: Wether to use ceil mode to calculate output height and with.
                       Defalut is True. If set false, Otherwise use floor.
 
-                      - ceil_mode=True:
-
-                      ..  math::
-
-                          w = 1 + int(ceil(input_width + 2 * padding - pool_size) / float(stride))
-                          h = 1 + int(ceil(input_height + 2 * padding_y - pool_size_y) / float(stride_y))
-
-                      - ceil_mode=False:
-
-                      ..  math::
-
-                          w = 1 + int(floor(input_width + 2 * padding - pool_size) / float(stride))
-                          h = 1 + int(floor(input_height + 2 * padding_y - pool_size_y) / float(stride_y))
     :type ceil_mode: bool
     :return: LayerOutput object.
     :rtype: LayerOutput
@@ -2199,6 +2243,15 @@ def spp_layer(input,
     The details please refer to
     `Kaiming He's paper <https://arxiv.org/abs/1406.4729>`_.
 
+    The example usage is:
+
+    ..  code-block:: python
+
+        spp = spp_layer(input=data, 
+                        pyramid_height=2, 
+                        num_channels=16, 
+                        pool_type=MaxPooling())
+
     :param name: layer name.
     :type name: basestring
     :param input: layer's input.
@@ -2287,6 +2340,12 @@ def img_cmrnorm_layer(input,
     The details please refer to
     `Alex's paper <http://www.cs.toronto.edu/~fritz/absps/imagenet.pdf>`_.
 
+    The example usage is:
+
+    ..  code-block:: python
+    
+        norm = img_cmrnorm_layer(input=net, size=5)
+
     :param name: layer name.
     :type name: None|basestring
     :param input: layer's input.
@@ -2342,6 +2401,12 @@ def batch_norm_layer(input,
     The details of batch normalization please refer to this
     `paper <http://arxiv.org/abs/1502.03167>`_.
 
+    The example usage is:
+
+    ..  code-block:: python
+    
+        norm = batch_norm_layer(input=net, act=ReluActivation())
+
     :param name: layer name.
     :type name: basestring
     :param input: batch normalization input. Better be linear activation.
@@ -3905,13 +3970,13 @@ def conv_shift_layer(a, b, name=None, layer_attr=None):
 
     .. code-block:: python
 
-       conv_shift = conv_shift_layer(input=[layer1, layer2])
+       conv_shift = conv_shift_layer(a=layer1, b=layer2)
 
     :param name: layer name
     :type name: basestring
     :param a: Input layer a.
     :type a: LayerOutput
-    :param b: input layer b
+    :param b: input layer b.
     :type b: LayerOutput
     :param layer_attr: layer's extra attribute.
     :type layer_attr: ExtraLayerAttribute
@@ -4003,8 +4068,8 @@ def tensor_layer(a,
 @wrap_act_default()
 @layer_support()
 def selective_fc_layer(input,
-                       select,
                        size,
+                       select=None,
                        act=None,
                        name=None,
                        pass_generation=False,
@@ -4031,6 +4096,7 @@ def selective_fc_layer(input,
     :type input: LayerOutput|list|tuple
     :param select: The select layer. The output of select layer should be a
                    sparse binary matrix, and treat as the mask of selective fc.
+                   If is None, acts exactly like fc_layer.
     :type select: LayerOutput
     :param size: The layer dimension.
     :type size: int
@@ -4259,7 +4325,7 @@ def block_expand_layer(input,
 
     .. code-block:: python
 
-       block_expand = block_expand_layer(input,
+       block_expand = block_expand_layer(input=layer,
                                          num_channels=128,
                                          stride_x=1,
                                          stride_y=1,
@@ -4463,7 +4529,7 @@ def warp_ctc_layer(input,
         - You can set 'blank' to any value ranged in [0, num_classes], which
           should be consistent as that used in your labels.
         - As a native 'softmax' activation is interated to the warp-ctc library,
-         'linear' activation is expected instead in the 'input' layer.
+          'linear' activation is expected instead in the 'input' layer.
 
     The simple usage:
 
@@ -4596,6 +4662,13 @@ def crf_decoding_layer(input,
     this layer will also calculate error. output.value[i] is 1 for incorrect
     decoding or 0 for correct decoding.
 
+    The simple usage:
+
+    .. code-block:: python
+
+      crf_decoding = crf_decoding_layer(input=input,
+                                        size=label_dim)
+
     :param input: The first input layer.
     :type input: LayerOutput
     :param size: size of this layer.

python/paddle/v2/__init__.py

@@ -18,13 +18,15 @@ import parameters
 import trainer
 import event
 import data_type
+import topology
 import data_feeder
 import attr
+import pooling
 import py_paddle.swig_paddle as api
 
 __all__ = [
     'optimizer', 'layer', 'activation', 'parameters', 'init', 'trainer',
-    'event', 'data_type', 'attr', 'data_feeder'
+    'event', 'data_type', 'attr', 'pooling', 'data_feeder', 'topology'
 ]
 
 

python/paddle/v2/data_feeder.py

@@ -23,7 +23,7 @@ class DataFeeder(DataProviderConverter):
     """
     DataFeeder converts the data returned by paddle.reader into a data structure
     of Arguments which is defined in the API. The paddle.reader usually returns
-    a list of mini-batch data entries. Each data entry in the list is one sampe.
+    a list of mini-batch data entries. Each data entry in the list is one sample.
     Each sample is a list or a tuple with one feature or multiple features.
     DataFeeder converts this mini-batch data entries into Arguments in order
     to feed it to C++ interface.

python/paddle/v2/data_type.py

@@ -13,10 +13,10 @@
 # limitations under the License.
 
 from paddle.trainer.PyDataProvider2 import \
-    InputType, dense_vector, sparse_binary_vector,\
+    InputType, DataType, dense_vector, sparse_binary_vector,\
     sparse_vector, integer_value, integer_value_sequence
 
 __all__ = [
-    'InputType', 'dense_vector', 'sparse_binary_vector', 'sparse_vector',
-    'integer_value', 'integer_value_sequence'
+    'InputType', 'DataType', 'dense_vector', 'sparse_binary_vector',
+    'sparse_vector', 'integer_value', 'integer_value_sequence'
 ]

python/paddle/v2/dataset/cifar.py

+"""
+CIFAR Dataset.
+
+URL: https://www.cs.toronto.edu/~kriz/cifar.html
+
+the default train_creator, test_creator used for CIFAR-10 dataset.
+"""
+import cPickle
+import itertools
+import tarfile
+
+import numpy
+
+from config import download
+
+__all__ = [
+    'cifar_100_train_creator', 'cifar_100_test_creator', 'train_creator',
+    'test_creator'
+]
+
+CIFAR10_URL = 'https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz'
+CIFAR10_MD5 = 'c58f30108f718f92721af3b95e74349a'
+CIFAR100_URL = 'https://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz'
+CIFAR100_MD5 = 'eb9058c3a382ffc7106e4002c42a8d85'
+
+
+def __read_batch__(filename, sub_name):
+    def reader():
+        def __read_one_batch_impl__(batch):
+            data = batch['data']
+            labels = batch.get('labels', batch.get('fine_labels', None))
+            assert labels is not None
+            for sample, label in itertools.izip(data, labels):
+                yield (sample / 255.0).astype(numpy.float32), int(label)
+
+        with tarfile.open(filename, mode='r') as f:
+            names = (each_item.name for each_item in f
+                     if sub_name in each_item.name)
+
+            for name in names:
+                batch = cPickle.load(f.extractfile(name))
+                for item in __read_one_batch_impl__(batch):
+                    yield item
+
+    return reader
+
+
+def cifar_100_train_creator():
+    fn = download(url=CIFAR100_URL, md5=CIFAR100_MD5)
+    return __read_batch__(fn, 'train')
+
+
+def cifar_100_test_creator():
+    fn = download(url=CIFAR100_URL, md5=CIFAR100_MD5)
+    return __read_batch__(fn, 'test')
+
+
+def train_creator():
+    """
+    Default train reader creator. Use CIFAR-10 dataset.
+    """
+    fn = download(url=CIFAR10_URL, md5=CIFAR10_MD5)
+    return __read_batch__(fn, 'data_batch')
+
+
+def test_creator():
+    """
+    Default test reader creator. Use CIFAR-10 dataset.
+    """
+    fn = download(url=CIFAR10_URL, md5=CIFAR10_MD5)
+    return __read_batch__(fn, 'test_batch')
+
+
+def unittest():
+    for _ in train_creator()():
+        pass
+    for _ in test_creator()():
+        pass
+
+
+if __name__ == '__main__':
+    unittest()

python/paddle/v2/dataset/config.py

+import hashlib
 import os
+import shutil
+import urllib2
 
-__all__ = ['DATA_HOME']
+__all__ = ['DATA_HOME', 'download']
 
 DATA_HOME = os.path.expanduser('~/.cache/paddle_data_set')
 
 if not os.path.exists(DATA_HOME):
     os.makedirs(DATA_HOME)
+
+
+def download(url, md5):
+    filename = os.path.split(url)[-1]
+    assert DATA_HOME is not None
+    filepath = os.path.join(DATA_HOME, md5)
+    if not os.path.exists(filepath):
+        os.makedirs(filepath)
+    __full_file__ = os.path.join(filepath, filename)
+
+    def __file_ok__():
+        if not os.path.exists(__full_file__):
+            return False
+        md5_hash = hashlib.md5()
+        with open(__full_file__, 'rb') as f:
+            for chunk in iter(lambda: f.read(4096), b""):
+                md5_hash.update(chunk)
+
+        return md5_hash.hexdigest() == md5
+
+    while not __file_ok__():
+        response = urllib2.urlopen(url)
+        with open(__full_file__, mode='wb') as of:
+            shutil.copyfileobj(fsrc=response, fdst=of)
+    return __full_file__

python/paddle/v2/dataset/movielens.py

+import zipfile
+from config import download
+import re
+import random
+import functools
+
+__all__ = ['train_creator', 'test_creator']
+
+
+class MovieInfo(object):
+    def __init__(self, index, categories, title):
+        self.index = int(index)
+        self.categories = categories
+        self.title = title
+
+    def value(self):
+        return [
+            self.index, [CATEGORIES_DICT[c] for c in self.categories],
+            [MOVIE_TITLE_DICT[w.lower()] for w in self.title.split()]
+        ]
+
+
+class UserInfo(object):
+    def __init__(self, index, gender, age, job_id):
+        self.index = int(index)
+        self.is_male = gender == 'M'
+        self.age = [1, 18, 25, 35, 45, 50, 56].index(int(age))
+        self.job_id = int(job_id)
+
+    def value(self):
+        return [self.index, 0 if self.is_male else 1, self.age, self.job_id]
+
+
+MOVIE_INFO = None
+MOVIE_TITLE_DICT = None
+CATEGORIES_DICT = None
+USER_INFO = None
+
+
+def __initialize_meta_info__():
+    fn = download(
+        url='http://files.grouplens.org/datasets/movielens/ml-1m.zip',
+        md5='c4d9eecfca2ab87c1945afe126590906')
+    global MOVIE_INFO
+    if MOVIE_INFO is None:
+        pattern = re.compile(r'^(.*)\((\d+)\)$')
+        with zipfile.ZipFile(file=fn) as package:
+            for info in package.infolist():
+                assert isinstance(info, zipfile.ZipInfo)
+                MOVIE_INFO = dict()
+                title_word_set = set()
+                categories_set = set()
+                with package.open('ml-1m/movies.dat') as movie_file:
+                    for i, line in enumerate(movie_file):
+                        movie_id, title, categories = line.strip().split('::')
+                        categories = categories.split('|')
+                        for c in categories:
+                            categories_set.add(c)
+                        title = pattern.match(title).group(1)
+                        MOVIE_INFO[int(movie_id)] = MovieInfo(
+                            index=movie_id, categories=categories, title=title)
+                        for w in title.split():
+                            title_word_set.add(w.lower())
+
+                global MOVIE_TITLE_DICT
+                MOVIE_TITLE_DICT = dict()
+                for i, w in enumerate(title_word_set):
+                    MOVIE_TITLE_DICT[w] = i
+
+                global CATEGORIES_DICT
+                CATEGORIES_DICT = dict()
+                for i, c in enumerate(categories_set):
+                    CATEGORIES_DICT[c] = i
+
+                global USER_INFO
+                USER_INFO = dict()
+                with package.open('ml-1m/users.dat') as user_file:
+                    for line in user_file:
+                        uid, gender, age, job, _ = line.strip().split("::")
+                        USER_INFO[int(uid)] = UserInfo(
+                            index=uid, gender=gender, age=age, job_id=job)
+    return fn
+
+
+def __reader__(rand_seed=0, test_ratio=0.1, is_test=False):
+    fn = __initialize_meta_info__()
+    rand = random.Random(x=rand_seed)
+    with zipfile.ZipFile(file=fn) as package:
+        with package.open('ml-1m/ratings.dat') as rating:
+            for line in rating:
+                if (rand.random() < test_ratio) == is_test:
+                    uid, mov_id, rating, _ = line.strip().split("::")
+                    uid = int(uid)
+                    mov_id = int(mov_id)
+                    rating = float(rating) * 2 - 5.0
+
+                    mov = MOVIE_INFO[mov_id]
+                    usr = USER_INFO[uid]
+                    yield usr.value() + mov.value() + [[rating]]
+
+
+def __reader_creator__(**kwargs):
+    return lambda: __reader__(**kwargs)
+
+
+train_creator = functools.partial(__reader_creator__, is_test=False)
+test_creator = functools.partial(__reader_creator__, is_test=True)
+
+
+def unittest():
+    for train_count, _ in enumerate(train_creator()()):
+        pass
+    for test_count, _ in enumerate(test_creator()()):
+        pass
+
+    print train_count, test_count
+
+
+if __name__ == '__main__':
+    unittest()

python/paddle/v2/layer.py

@@ -82,10 +82,17 @@ import activation
 import attr
 
 __all__ = [
-    'parse_network', 'data', 'fc', 'max_id', 'classification_cost',
-    'cross_entropy_cost', 'cross_entropy_with_selfnorm_cost', 'regression_cost',
+    'parse_network', 'data', 'fc', 'conv_shift', 'img_conv', 'img_pool', 'spp',
+    'maxout', 'img_cmrnorm', 'batch_norm', 'sum_to_one_norm', 'recurrent',
+    'lstmemory', 'grumemory', 'pool', 'last_seq', 'first_seq', 'concat',
+    'seq_concat', 'block_expand', 'expand', 'repeat', 'seq_reshape', 'addto',
+    'linear_comb', 'interpolation', 'bilinear_interp', 'power', 'scaling',
+    'slope_intercept', 'tensor', 'cos_sim', 'trans', 'max_id', 'sampling_id',
+    'pad', 'classification_cost', 'cross_entropy_cost',
+    'cross_entropy_with_selfnorm_cost', 'regression_cost',
     'multi_binary_label_cross_entropy_cost', 'rank_cost', 'lambda_cost',
-    'sum_cost', 'huber_cost'
+    'sum_cost', 'huber_cost', 'crf', 'crf_decoding', 'ctc', 'warp_ctc', 'nce',
+    'hsigmoid', 'eos'
 ]
 
 __projection_names__ = filter(lambda x: x.endswith('_projection'),
@@ -143,9 +150,9 @@ class Layer(object):
         raise NotImplementedError()
 
 
-def __convert_to_v2__(method_name, name_prefix=None, parent_names=None):
-    if name_prefix is not None:
-        wrapper = wrap_name_default(name_prefix=name_prefix)
+def __convert_to_v2__(method_name, parent_names, is_default_name=True):
+    if is_default_name:
+        wrapper = wrap_name_default(name_prefix=method_name)
     else:
         wrapper = None
 
@@ -277,44 +284,93 @@ def mixed(size=0,
     return MixedLayerV2(size, input, name, act, bias_attr, layer_attr)
 
 
+LayerV2 = Layer
 data = DataLayerV2
-fc = __convert_to_v2__('fc_layer', name_prefix='fc', parent_names=['input'])
-max_id = __convert_to_v2__(
-    'maxid_layer', name_prefix='maxid', parent_names=['input'])
-classification_cost = __convert_to_v2__(
-    'classification_cost',
-    name_prefix='classification_cost',
-    parent_names=['input', 'label', 'weight'])
-regression_cost = __convert_to_v2__(
-    'regression_cost',
-    name_prefix='regression_cost',
-    parent_names=['input', 'label', 'weight'])
-cross_entropy_cost = __convert_to_v2__(
-    'cross_entropy',
-    name_prefix='cross_entropy',
-    parent_names=['input', 'label'])
-cross_entropy_with_selfnorm_cost = __convert_to_v2__(
-    'cross_entropy_with_selfnorm',
-    name_prefix='cross_entropy_with_selfnorm',
-    parent_names=['input', 'label'])
-multi_binary_label_cross_entropy_cost = __convert_to_v2__(
-    'multi_binary_label_cross_entropy',
-    name_prefix='multi_binary_label_cross_entropy',
-    parent_names=['input', 'label'])
-rank_cost = __convert_to_v2__(
-    'rank_cost',
-    name_prefix='rank_cost',
-    parent_names=['left', 'right', 'label', 'weight'])
-lambda_cost = __convert_to_v2__(
-    'lambda_cost', name_prefix='lambda_cost', parent_names=['input', 'score'])
-sum_cost = __convert_to_v2__(
-    'sum_cost', name_prefix='sum_cost', parent_names=['input'])
-huber_cost = __convert_to_v2__(
-    'huber_cost', name_prefix='huber_cost', parent_names=['input', 'label'])
+AggregateLevel = conf_helps.layers.AggregateLevel
+ExpandLevel = conf_helps.layers.ExpandLevel
+
+layer_list = [
+    # [V2LayerImpl, V1_method_name, parent_names]
+    # fully connected layers
+    ['fc', 'fc_layer', ['input']],
+    # conv layers
+    ['conv_shift', 'conv_shift_layer', ['a', 'b']],
+    ['img_conv', 'img_conv_layer', ['input']],
+    # image pooling layers
+    ['img_pool', 'img_pool_layer', ['input']],
+    ['spp', 'spp_layer', ['input']],
+    ['maxout', 'maxout_layer', ['input']],
+    # norm layers
+    ['img_cmrnorm', 'img_cmrnorm_layer', ['input']],
+    ['batch_norm', 'batch_norm_layer', ['input']],
+    ['sum_to_one_norm', 'sum_to_one_norm_layer', ['input']],
+    # recurrent layers
+    ['recurrent', 'recurrent_layer', ['input']],
+    ['lstmemory', 'lstmemory', ['input']],
+    ['grumemory', 'grumemory', ['input']],
+    # aggregate layers
+    ['pool', 'pooling_layer', ['input']],
+    ['last_seq', 'last_seq', ['input']],
+    ['first_seq', 'first_seq', ['input']],
+    ['concat', 'concat_layer', ['input']],
+    ['seq_concat', 'seq_concat_layer', ['a', 'b']],
+    # reshaping layers
+    ['block_expand', 'block_expand_layer', ['input']],
+    ['expand', 'expand_layer', ['input', 'expand_as']],
+    ['repeat', 'repeat_layer', ['input']],
+    ['rotate', 'rotate_layer', ['input']],
+    ['seq_reshape', 'seq_reshape_layer', ['input']],
+    # math layers
+    ['addto', 'addto_layer', ['input']],
+    ['linear_comb', 'linear_comb_layer', ['weights', 'vectors']],
+    ['interpolation', 'interpolation_layer', ['input', 'weight']],
+    ['bilinear_interp', 'bilinear_interp_layer', ['input']],
+    ['power', 'power_layer', ['input', 'weight']],
+    ['scaling', 'scaling_layer', ['input', 'weight']],
+    ['slope_intercept', 'slope_intercept_layer', ['input']],
+    ['tensor', 'tensor_layer', ['a', 'b']],
+    ['cos_sim', 'cos_sim', ['a', 'b']],
+    ['trans', 'trans_layer', ['input']],
+    # sampling layers
+    ['max_id', 'maxid_layer', ['input']],
+    ['sampling_id', 'sampling_id_layer', ['input']],
+    # slicing and joining layers
+    ['pad', 'pad_layer', ['input']],
+    # cost layers
+    [
+        'classification_cost', 'classification_cost',
+        ['input', 'label', 'weight']
+    ],
+    ['regression_cost', 'regression_cost', ['input', 'label', 'weight']],
+    ['cross_entropy_cost', 'cross_entropy', ['input', 'label']],
+    [
+        'cross_entropy_with_selfnorm_cost', 'cross_entropy_with_selfnorm',
+        ['input', 'label']
+    ],
+    [
+        'multi_binary_label_cross_entropy_cost',
+        'multi_binary_label_cross_entropy', ['input', 'label']
+    ],
+    ['rank_cost', 'rank_cost', ['left', 'right', 'label', 'weight']],
+    ['lambda_cost', 'lambda_cost', ['input', 'score']],
+    ['sum_cost', 'sum_cost', ['input']],
+    ['huber_cost', 'huber_cost', ['input', 'label']],
+    ['crf', 'crf_layer', ['input', 'label']],
+    ['crf_decoding', 'crf_decoding_layer', ['input']],
+    ['ctc', 'ctc_layer', ['input', 'label']],
+    ['warp_ctc', 'warp_ctc_layer', ['input', 'label']],
+    ['nce', 'nce_layer', ['input', 'label']],
+    ['hsigmoid', 'hsigmoid', ['input', 'label']],
+    # check layers
+    ['eos', 'eos_layer', ['input']]
+]
+for l in layer_list:
+    globals()[l[0]] = __convert_to_v2__(l[1], l[2])
 
 # convert projection
 for prj in __projection_names__:
-    globals()[prj] = __convert_to_v2__(prj, parent_names=['input'])
+    globals()[prj] = __convert_to_v2__(
+        prj, parent_names=['input'], is_default_name=False)
 
 # convert operator
 operator_list = [
@@ -323,4 +379,5 @@ operator_list = [
     ['conv_operator', ['img', 'filter']]
 ]
 for op in operator_list:
-    globals()[op[0]] = __convert_to_v2__(op[0], parent_names=op[1])
+    globals()[op[0]] = __convert_to_v2__(
+        op[0], parent_names=op[1], is_default_name=False)

python/paddle/v2/parameters.py

 import numpy as np
-from . import layer as v2_layer
 import py_paddle.swig_paddle as api
 from paddle.proto.ParameterConfig_pb2 import ParameterConfig
 
+from topology import Topology
+
 __all__ = ['Parameters', 'create']
 
 
-def create(*layers):
+def create(layers):
     """
-    Create parameter pool by layers. In paddle, layer can be represent a
-    model config.
-
+    Create parameter pool by topology.
     :param layers:
     :return:
     """
-    for layer in layers:
-        if not isinstance(layer, v2_layer.Layer):
-            raise ValueError(
-                'create must pass a topologies which type is paddle.layer.Layer')
-    model_config = v2_layer.parse_network(*layers)
+    topology = Topology(layers)
     pool = Parameters()
-    for param in model_config.parameters:
+    for param in topology.proto().parameters:
         pool.__append_config__(param)
     return pool
 
@@ -224,7 +219,8 @@ class Parameters(object):
                 except ValueError:
                     # If no such parameter in gradient machine, then don't copy
                     pass
-            self.__gradient_machines__.append(gradient_machine)
+
+        self.__gradient_machines__.append(gradient_machine)
 
 
 def __get_parameter_in_gradient_machine__(gradient_machine, name):

python/paddle/v2/pooling.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.
+
+from paddle.trainer_config_helpers.poolings import *
+
+__all__ = ["Max", "CudnnMax", "Avg", "CudnnAvg", "Sum", "SquareRootN"]
+
+Max = MaxPooling
+CudnnMax = CudnnMaxPooling
+Avg = AvgPooling
+CudnnAvg = CudnnAvgPooling
+Sum = SumPooling
+SquareRootN = SquareRootNPooling

python/paddle/v2/tests/CMakeLists.txt

@@ -2,5 +2,11 @@ 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})
+        COMMAND bash ${PROJ_ROOT}/python/paddle/v2/tests/run_tests.sh ${PYTHON_EXECUTABLE})
+
+add_test(NAME topology_test
+        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_layer.py

@@ -19,16 +19,106 @@ 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.pooling as pooling
+from paddle.trainer_config_helpers.config_parser_utils import \
+    parse_network_config as parse_network
 
-pixel = layer.data(name='pixel', type=data_type.dense_vector(784))
+pixel = layer.data(name='pixel', type=data_type.dense_vector(128))
 label = layer.data(name='label', type=data_type.integer_value(10))
 weight = layer.data(name='weight', type=data_type.dense_vector(10))
 score = layer.data(name='score', type=data_type.dense_vector(1))
+
 hidden = layer.fc(input=pixel,
                   size=100,
                   act=activation.Sigmoid(),
                   param_attr=attr.Param(name='hidden'))
 inference = layer.fc(input=hidden, size=10, act=activation.Softmax())
+conv = layer.img_conv(
+    input=pixel,
+    filter_size=1,
+    filter_size_y=1,
+    num_channels=8,
+    num_filters=16,
+    act=activation.Linear())
+
+
+class ImageLayerTest(unittest.TestCase):
+    def test_conv_layer(self):
+        conv_shift = layer.conv_shift(a=pixel, b=score)
+        print layer.parse_network(conv, conv_shift)
+
+    def test_pooling_layer(self):
+        maxpool = layer.img_pool(
+            input=conv,
+            pool_size=2,
+            num_channels=16,
+            padding=1,
+            pool_type=pooling.Max())
+        spp = layer.spp(input=conv,
+                        pyramid_height=2,
+                        num_channels=16,
+                        pool_type=pooling.Max())
+        maxout = layer.maxout(input=conv, num_channels=16, groups=4)
+        print layer.parse_network(maxpool, spp, maxout)
+
+    def test_norm_layer(self):
+        norm1 = layer.img_cmrnorm(input=conv, size=5)
+        norm2 = layer.batch_norm(input=conv)
+        norm3 = layer.sum_to_one_norm(input=conv)
+        print layer.parse_network(norm1, norm2, norm3)
+
+
+class AggregateLayerTest(unittest.TestCase):
+    def test_aggregate_layer(self):
+        pool = layer.pool(
+            input=pixel,
+            pooling_type=pooling.Avg(),
+            agg_level=layer.AggregateLevel.EACH_SEQUENCE)
+        last_seq = layer.last_seq(input=pixel)
+        first_seq = layer.first_seq(input=pixel)
+        concat = layer.concat(input=[last_seq, first_seq])
+        seq_concat = layer.seq_concat(a=last_seq, b=first_seq)
+        print layer.parse_network(pool, last_seq, first_seq, concat, seq_concat)
+
+
+class MathLayerTest(unittest.TestCase):
+    def test_math_layer(self):
+        addto = layer.addto(input=[pixel, pixel])
+        linear_comb = layer.linear_comb(weights=weight, vectors=hidden, size=10)
+        interpolation = layer.interpolation(
+            input=[hidden, hidden], weight=score)
+        bilinear = layer.bilinear_interp(input=conv, out_size_x=4, out_size_y=4)
+        power = layer.power(input=pixel, weight=score)
+        scaling = layer.scaling(input=pixel, weight=score)
+        slope = layer.slope_intercept(input=pixel)
+        tensor = layer.tensor(a=pixel, b=pixel, size=1000)
+        cos_sim = layer.cos_sim(a=pixel, b=pixel)
+        trans = layer.trans(input=tensor)
+        print layer.parse_network(addto, linear_comb, interpolation, power,
+                                  scaling, slope, tensor, cos_sim, trans)
+
+
+class ReshapeLayerTest(unittest.TestCase):
+    def test_reshape_layer(self):
+        block_expand = layer.block_expand(
+            input=conv, num_channels=4, stride_x=1, block_x=1)
+        expand = layer.expand(
+            input=weight,
+            expand_as=pixel,
+            expand_level=layer.ExpandLevel.FROM_TIMESTEP)
+        repeat = layer.repeat(input=pixel, num_repeats=4)
+        reshape = layer.seq_reshape(input=pixel, reshape_size=4)
+        rotate = layer.rotate(input=pixel, height=16, width=49)
+        print layer.parse_network(block_expand, expand, repeat, reshape, rotate)
+
+
+class RecurrentLayerTest(unittest.TestCase):
+    def test_recurrent_layer(self):
+        word = layer.data(name='word', type=data_type.integer_value(12))
+        recurrent = layer.recurrent(input=word)
+        lstm = layer.lstmemory(input=word)
+        gru = layer.grumemory(input=word)
+        print layer.parse_network(recurrent, lstm, gru)
 
 
 class CostLayerTest(unittest.TestCase):
@@ -49,13 +139,35 @@ class CostLayerTest(unittest.TestCase):
         cost10 = layer.sum_cost(input=inference)
         cost11 = layer.huber_cost(input=score, label=label)
 
-        print dir(layer)
-        layer.parse_network(cost1, cost2)
-        print dir(layer)
-        #print layer.parse_network(cost3, cost4)
-        #print layer.parse_network(cost5, cost6)
-        #print layer.parse_network(cost7, cost8, cost9, cost10, cost11)
+        print layer.parse_network(cost1, cost2)
+        print layer.parse_network(cost3, cost4)
+        print layer.parse_network(cost5, cost6)
+        print layer.parse_network(cost7, cost8, cost9, cost10, cost11)
+
+        crf = layer.crf(input=inference, label=label)
+        crf_decoding = layer.crf_decoding(input=inference, size=3)
+        ctc = layer.ctc(input=inference, label=label)
+        warp_ctc = layer.warp_ctc(input=pixel, label=label)
+        nce = layer.nce(input=inference, label=label, num_classes=3)
+        hsigmoid = layer.hsigmoid(input=inference, label=label, num_classes=3)
+
+        print layer.parse_network(crf, crf_decoding, ctc, warp_ctc, nce,
+                                  hsigmoid)
+
+
+class OtherLayerTest(unittest.TestCase):
+    def test_sampling_layer(self):
+        maxid = layer.max_id(input=inference)
+        sampling_id = layer.sampling_id(input=inference)
+        eos = layer.eos(input=maxid, eos_id=5)
+        print layer.parse_network(maxid, sampling_id, eos)
+
+    def test_slicing_joining_layer(self):
+        pad = layer.pad(input=conv, pad_c=[2, 3], pad_h=[1, 2], pad_w=[3, 1])
+        print layer.parse_network(pad)
+
 
+class ProjOpTest(unittest.TestCase):
     def test_projection(self):
         input = layer.data(name='data', type=data_type.dense_vector(784))
         word = layer.data(

python/paddle/v2/tests/test_topology.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 unittest
+import paddle.v2.layer as layer
+import paddle.v2.topology as topology
+import paddle.v2.data_type as data_type
+import paddle.trainer_config_helpers as conf_helps
+
+
+class TestTopology(unittest.TestCase):
+    def test_data_type(self):
+        pixel = layer.data(name='pixel', type=data_type.dense_vector(784))
+        label = layer.data(name='label', type=data_type.integer_value(10))
+        hidden = layer.fc(input=pixel,
+                          size=100,
+                          act=conf_helps.SigmoidActivation())
+        inference = layer.fc(input=hidden,
+                             size=10,
+                             act=conf_helps.SoftmaxActivation())
+        cost = layer.classification_cost(input=inference, label=label)
+        topo = topology.Topology(cost)
+        data_types = topo.data_type()
+        self.assertEqual(len(data_types), 2)
+        pixel_data_type = filter(lambda type: type[0] == "pixel", data_types)
+        self.assertEqual(len(pixel_data_type), 1)
+        pixel_data_type = pixel_data_type[0]
+        self.assertEqual(pixel_data_type[1].type, data_type.DataType.Dense)
+        self.assertEqual(pixel_data_type[1].dim, 784)
+
+        label_data_type = filter(lambda type: type[0] == "label", data_types)
+        self.assertEqual(len(label_data_type), 1)
+        label_data_type = label_data_type[0]
+        self.assertEqual(label_data_type[1].type, data_type.DataType.Index)
+        self.assertEqual(label_data_type[1].dim, 10)
+
+    def test_get_layer(self):
+        pixel = layer.data(name='pixel', type=data_type.dense_vector(784))
+        label = layer.data(name='label', type=data_type.integer_value(10))
+        hidden = layer.fc(input=pixel,
+                          size=100,
+                          act=conf_helps.SigmoidActivation())
+        inference = layer.fc(input=hidden,
+                             size=10,
+                             act=conf_helps.SoftmaxActivation())
+        cost = layer.classification_cost(input=inference, label=label)
+        topo = topology.Topology(cost)
+        pixel_layer = topo.get_layer("pixel")
+        label_layer = topo.get_layer("label")
+        self.assertEqual(pixel_layer, pixel)
+        self.assertEqual(label_layer, label)
+
+    def test_parse(self):
+        pixel = layer.data(name='pixel', type=data_type.dense_vector(784))
+        label = layer.data(name='label', type=data_type.integer_value(10))
+        hidden = layer.fc(input=pixel,
+                          size=100,
+                          act=conf_helps.SigmoidActivation())
+        inference = layer.fc(input=hidden,
+                             size=10,
+                             act=conf_helps.SoftmaxActivation())
+        maxid = layer.max_id(input=inference)
+        cost1 = layer.classification_cost(input=inference, label=label)
+        cost2 = layer.cross_entropy_cost(input=inference, label=label)
+
+        topology.Topology(cost2).proto()
+        topology.Topology([cost1]).proto()
+        topology.Topology([cost1, cost2]).proto()
+        topology.Topology([inference, maxid]).proto()
+
+
+if __name__ == '__main__':
+    unittest.main()

python/paddle/v2/topology.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 collections
+
+from paddle.proto.ModelConfig_pb2 import ModelConfig
+
+import layer as v2_layer
+
+__all__ = ['Topology']
+
+
+class Topology(object):
+    """
+    Topology is used to store the information about all layers
+    and network configs.
+    """
+
+    def __init__(self, layers):
+        if not isinstance(layers, collections.Sequence):
+            __check_layer_type__(layers)
+            layers = [layers]
+        for layer in layers:
+            __check_layer_type__(layer)
+        self.layers = layers
+        self.__model_config__ = v2_layer.parse_network(*layers)
+        assert isinstance(self.__model_config__, ModelConfig)
+
+    def proto(self):
+        return self.__model_config__
+
+    def get_layer(self, name):
+        """
+        get v2.Layer Class instance by layer name
+        :param name:
+        :return:
+        """
+        result_layer = []
+
+        def find_layer_by_name(layer, layer_name):
+            if len(result_layer) == 1:
+                return
+            elif layer.name == layer_name:
+                result_layer.append(layer)
+            else:
+                for parent_layer in layer.__parent_layers__.values():
+                    find_layer_by_name(parent_layer, layer_name)
+
+        for layer in self.layers:
+            find_layer_by_name(layer, name)
+
+        assert len(result_layer) == 1
+        return result_layer[0]
+
+    def data_layers(self):
+        """
+        get all data layer
+        :return:
+        """
+        data_layers = set()
+
+        def find_data_layer(layer):
+            if isinstance(layer, v2_layer.DataLayerV2):
+                data_layers.add(layer)
+            for parent_layer in layer.__parent_layers__.values():
+                find_data_layer(parent_layer)
+
+        for layer in self.layers:
+            find_data_layer(layer)
+
+        return data_layers
+
+    def data_type(self):
+        """
+        get data_type from proto, such as:
+        [('image', dense_vector(768)), ('label', integer_value(10))]
+        """
+        return [(data_layer.name, data_layer.type)
+                for data_layer in self.data_layers()]
+
+
+def __check_layer_type__(layer):
+    if not isinstance(layer, v2_layer.LayerV2):
+        raise ValueError('layer should have type paddle.layer.Layer')

python/paddle/v2/trainer.py

 import collections
 
 import py_paddle.swig_paddle as api
-from paddle.proto.ModelConfig_pb2 import ModelConfig
-from data_feeder import DataFeeder
 
+from data_feeder import DataFeeder
+from topology import Topology
 from . import event as v2_event
-from . import layer as v2_layer
 from . import optimizer as v2_optimizer
 from . import parameters as v2_parameters
 
@@ -30,7 +29,7 @@ class ITrainer(object):
 
     def train(self,
               train_data_reader,
-              topology,
+              cost,
               parameters,
               test_data_reader=None,
               event_handler=None):
@@ -38,7 +37,7 @@ class ITrainer(object):
         train method.
 
         :param train_data_reader:
-        :param topology:
+        :param cost:
         :param parameters:
         :param test_data_reader:
         :param event_handler:
@@ -63,19 +62,18 @@ class SGD(ITrainer):
 
     def train(self,
               train_data_reader,
-              topology,
+              cost,
               parameters,
               num_passes=1,
               test_data_reader=None,
               event_handler=None,
               batch_size=32,
-              data_types=None,
               reader_dict=None):
         """
         Training method. Will train num_passes of input data.
 
         :param train_data_reader:
-        :param topology: Network Topology, use one or more Layers to represent it.
+        :param cost: cost layers, to be optimized.
         :param parameters: The parameter pools.
         :param num_passes: The total train passes.
         :param test_data_reader:
@@ -83,18 +81,18 @@ class SGD(ITrainer):
                               occurred.
         :type event_handler: (BaseEvent) => None
         :param batch_size: Not important, will be removed after data refactor.
-        :param data_types: Not important, will be removed after data refactor.
         :return:
         """
         if event_handler is None:
             event_handler = default_event_handler
 
-        topology = v2_layer.parse_network(topology)
+        topology = Topology(cost)
 
         __check_train_args__(**locals())
 
         gm = api.GradientMachine.createFromConfigProto(
-            topology, api.CREATE_MODE_NORMAL, self.__optimizer__.enable_types())
+            topology.proto(), api.CREATE_MODE_NORMAL,
+            self.__optimizer__.enable_types())
         assert isinstance(gm, api.GradientMachine)
         parameters.append_gradient_machine(gm)
         gm.randParameters()
@@ -108,7 +106,7 @@ class SGD(ITrainer):
         assert isinstance(pass_evaluator, api.Evaluator)
         out_args = api.Arguments.createArguments(0)
 
-        feeder = DataFeeder(data_types, reader_dict)
+        feeder = DataFeeder(topology.data_type(), reader_dict)
 
         for pass_id in xrange(num_passes):
             event_handler(v2_event.BeginPass(pass_id))
@@ -154,7 +152,7 @@ def __data_reader_to_batch__(reader, batch_size, topology):
     def input_reorder(func):
         for item in func():
             retv = []
-            for __layer_name__ in topology.input_layer_names:
+            for __layer_name__ in topology.proto().input_layer_names:
                 retv.append(item[__layer_name__])
             yield retv
 
@@ -191,7 +189,7 @@ def __check_train_args__(train_data_reader, topology, parameters,
             raise ValueError('test_data_reader should be a function, which can '
                              'return a iterator')
 
-    if not isinstance(topology, ModelConfig):
+    if not isinstance(topology, Topology):
         raise ValueError('topology should be a model config')
 
     if not isinstance(parameters, v2_parameters.Parameters):