Remove the use of global variables. (#388)

luomai · zsdonghao · commit 430397930ff2 · 2018-03-09T16:48:01.000Z
* fixed bug of time distributed layer and release 2 tests

* speed up test mnist

* fixed test file name

* remove unused code
diff --git a/.travis.yml b/.travis.yml
@@ -20,3 +20,5 @@ script:
   - python tests/test_pydocstyle.py
   - python tests/test_mnist_simple.py
   - python tests/test_pooling.py
+  - python tests/test_reuse_mlp.py
+  - python tests/test_time_distributed.py
diff --git a/example/tutorial_cifar10.py b/example/tutorial_cifar10.py
@@ -21,7 +21,6 @@ def model(x, y_, reuse):
     W_init2 = tf.truncated_normal_initializer(stddev=0.04)
     b_init2 = tf.constant_initializer(value=0.1)
     with tf.variable_scope("model", reuse=reuse):
-        tl.layers.set_name_reuse(reuse)
         net = InputLayer(x, name='input')
         net = Conv2d(net, 64, (5, 5), (1, 1), act=tf.nn.relu, padding='SAME', W_init=W_init, name='cnn1')
         # net = Conv2dLayer(net, act=tf.nn.relu, shape=[5, 5, 3, 64],
@@ -69,7 +68,6 @@ def model_batch_norm(x, y_, reuse, is_train):
     W_init2 = tf.truncated_normal_initializer(stddev=0.04)
     b_init2 = tf.constant_initializer(value=0.1)
     with tf.variable_scope("model", reuse=reuse):
-        tl.layers.set_name_reuse(reuse)
         net = InputLayer(x, name='input')
 
         net = Conv2d(net, 64, (5, 5), (1, 1), padding='SAME', W_init=W_init, b_init=None, name='cnn1')
diff --git a/example/tutorial_cifar10_tfrecord.py b/example/tutorial_cifar10_tfrecord.py
@@ -44,6 +44,7 @@
 
 # import numpy as np
 import tensorflow as tf
+
 import tensorlayer as tl
 from tensorlayer.layers import *
 
@@ -185,7 +186,6 @@ def model(x_crop, y_, reuse):
         W_init2 = tf.truncated_normal_initializer(stddev=0.04)
         b_init2 = tf.constant_initializer(value=0.1)
         with tf.variable_scope("model", reuse=reuse):
-            tl.layers.set_name_reuse(reuse)
             net = InputLayer(x_crop, name='input')
             net = Conv2d(net, 64, (5, 5), (1, 1), act=tf.nn.relu, padding='SAME', W_init=W_init, name='cnn1')
             # net = Conv2dLayer(net, act=tf.nn.relu, shape=[5, 5, 3, 64],
@@ -233,7 +233,6 @@ def model_batch_norm(x_crop, y_, reuse, is_train):
         W_init2 = tf.truncated_normal_initializer(stddev=0.04)
         b_init2 = tf.constant_initializer(value=0.1)
         with tf.variable_scope("model", reuse=reuse):
-            tl.layers.set_name_reuse(reuse)
             net = InputLayer(x_crop, name='input')
 
             net = Conv2d(net, 64, (5, 5), (1, 1), padding='SAME', W_init=W_init, b_init=None, name='cnn1')
diff --git a/example/tutorial_generate_text.py b/example/tutorial_generate_text.py
@@ -231,7 +231,6 @@ def inference(x, is_train, sequence_length, reuse=None):
         print("\nsequence_length: %d, is_train: %s, reuse: %s" % (sequence_length, is_train, reuse))
         rnn_init = tf.random_uniform_initializer(-init_scale, init_scale)
         with tf.variable_scope("model", reuse=reuse):
-            tl.layers.set_name_reuse(reuse)
             network = EmbeddingInputlayer(inputs=x, vocabulary_size=vocab_size, embedding_size=hidden_size, E_init=rnn_init, name='embedding')
             network = RNNLayer(
                 network,
diff --git a/example/tutorial_mlp_dropout2.py b/example/tutorial_mlp_dropout2.py
@@ -18,7 +18,6 @@
 # define the network
 def mlp(x, is_train=True, reuse=False):
     with tf.variable_scope("MLP", reuse=reuse):
-        tl.layers.set_name_reuse(reuse)
         network = tl.layers.InputLayer(x, name='input')
         network = tl.layers.DropoutLayer(network, keep=0.8, is_fix=True, is_train=is_train, name='drop1')
         network = tl.layers.DenseLayer(network, n_units=800, act=tf.nn.relu, name='relu1')
diff --git a/example/tutorial_mnist_float16.py b/example/tutorial_mnist_float16.py
@@ -20,7 +20,6 @@
 
 def model(x, is_train=True, reuse=False):
     with tf.variable_scope("model", reuse=reuse):
-        tl.layers.set_name_reuse(reuse)
         n = InputLayer(x, name='input')
         # cnn
         n = Conv2d(n, 32, (5, 5), (1, 1), padding='SAME', name='cnn1')
diff --git a/example/tutorial_ptb_lstm.py b/example/tutorial_ptb_lstm.py
@@ -194,7 +194,6 @@ def inference(x, is_training, num_steps, reuse=None):
         print("\nnum_steps : %d, is_training : %s, reuse : %s" % (num_steps, is_training, reuse))
         initializer = tf.random_uniform_initializer(-init_scale, init_scale)
         with tf.variable_scope("model", reuse=reuse):
-            tl.layers.set_name_reuse(reuse)
             network = tl.layers.EmbeddingInputlayer(inputs=x, vocabulary_size=vocab_size, embedding_size=hidden_size, E_init=initializer, name='embedding')
             network = tl.layers.DropoutLayer(network, keep=keep_prob, is_fix=True, is_train=is_training, name='drop1')
             network = tl.layers.RNNLayer(
diff --git a/example/tutorial_ptb_lstm_state_is_tuple.py b/example/tutorial_ptb_lstm_state_is_tuple.py
@@ -194,7 +194,6 @@ def inference(x, is_training, num_steps, reuse=None):
         print("\nnum_steps : %d, is_training : %s, reuse : %s" % (num_steps, is_training, reuse))
         initializer = tf.random_uniform_initializer(-init_scale, init_scale)
         with tf.variable_scope("model", reuse=reuse):
-            tl.layers.set_name_reuse(reuse)
             network = tl.layers.EmbeddingInputlayer(inputs=x, vocabulary_size=vocab_size, embedding_size=hidden_size, E_init=initializer, name='embedding')
             network = tl.layers.DropoutLayer(network, keep=keep_prob, is_fix=True, is_train=is_training, name='drop1')
             network = tl.layers.RNNLayer(
diff --git a/tensorlayer/layers/core.py b/tensorlayer/layers/core.py
@@ -4,6 +4,7 @@
 
 import numpy as np
 import tensorflow as tf
+from tensorflow.python.util.deprecation import deprecated
 
 from .. import _logging as logging
 from .. import files, iterate, utils, visualize
@@ -12,8 +13,6 @@
 class LayersConfig:
     tf_dtype = tf.float32  # TensorFlow DType
     set_keep = {}  # A dictionary for holding tf.placeholders
-    _layer_name_list = []  # A list of used layer names
-    _name_reuse = False  # Boolean to indicate if layer names can be reused
 
 
 try:  # For TF12 and later
@@ -60,73 +59,14 @@ def flatten_reshape(variable, name='flatten'):
     return tf.reshape(variable, shape=[-1, dim], name=name)
 
 
+@deprecated("2018-06-30", "TensorLayer relies on TensorFlow to check naming.")
 def clear_layers_name():
-    """Clear all layer names in LayersConfig
-
-    Examples
-    ---------
-    Clean the current graph and try to re-define model.
-
-    >>> for .... (different model settings):
-    >>>    with tf.Graph().as_default() as graph:   # clear all variables of TF
-    >>>       tl.layers.clear_layers_name()         # clear all layer name of TL
-    >>>       sess = tf.InteractiveSession()
-    >>>       # define and train a model here
-    >>>       sess.close()
-
-    Enable reusing layer names.
-
-    >>> net = tl.layers.InputLayer(x, name='input_layer')
-    >>> net = tl.layers.DenseLayer(net, n_units=800, name='relu1')
-    ...
-    >>> tl.layers.clear_layers_name()
-    >>> net2 = tl.layers.InputLayer(x, name='input_layer')
-    >>> net2 = tl.layers.DenseLayer(net2, n_units=800, name='relu1')
-
-    """
-    LayersConfig._layer_name_list = []
+    logging.warning('this method is DEPRECATED and has no effect, please remove it from your code.')
 
 
+@deprecated("2018-06-30", "TensorLayer relies on TensorFlow to check name reusing.")
 def set_name_reuse(enable=True):
-    """Enable or disable reuse layer name.
-
-    By default, each layer must has unique
-    name. When you want two or more input placeholder (inference) share the same
-    model parameters, you need to enable layer name reuse, then allow the
-    parameters have same name scope.
-
-    Parameters
-    ----------
-    enable : boolean
-        Enable or disable name/layer reuse, None means False.
-
-    Examples
-    --------
-    >>> def embed_seq(input_seqs, is_train, reuse):
-    >>>    with tf.variable_scope("model", reuse=reuse):
-    >>>         tl.layers.set_name_reuse(reuse)
-    >>>         net = tl.layers.EmbeddingInputlayer(
-    ...                     inputs = input_seqs,
-    ...                     vocabulary_size = vocab_size,
-    ...                     embedding_size = embedding_size,
-    ...                     name = 'e_embedding')
-    >>>         net = tl.layers.DynamicRNNLayer(net,
-    ...                     cell_fn = tf.contrib.rnn.BasicLSTMCell,
-    ...                     n_hidden = embedding_size,
-    ...                     dropout = (0.7 if is_train else None),
-    ...                     initializer = w_init,
-    ...                     sequence_length = tl.layers.retrieve_seq_length_op2(input_seqs),
-    ...                     return_last = True,
-    ...                     name = 'e_dynamicrnn')
-    >>>    return net
-    >>>
-    >>> net_train = embed_seq(t_caption, is_train=True, reuse=False)
-    >>> net_test = embed_seq(t_caption, is_train=False, reuse=True)
-
-    - see ``tutorial_ptb_lstm.py`` for example.
-
-    """
-    LayersConfig._name_reuse = enable
+    logging.warning('this method is DEPRECATED and has no effect, please remove it from your code.')
 
 
 def initialize_rnn_state(state, feed_dict=None):
@@ -383,15 +323,7 @@ def __init__(self, inputs=None, name='layer'):
         scope_name = tf.get_variable_scope().name
         if scope_name:
             name = scope_name + '/' + name
-        if (name in LayersConfig._layer_name_list) and LayersConfig._name_reuse is False:
-            raise Exception("Layer '%s' already exists, please choice other 'name' or reuse this layer\
-            \nHint : Use different name for different 'Layer' (The name is used to control parameter sharing)\
-            \nAdditional Informations: http://tensorlayer.readthedocs.io/en/latest/modules/layers.html?highlight=clear_layers_name#tensorlayer.layers.clear_layers_name"
-                            % name)
-        else:
-            self.name = name
-            if name not in ['', None, False]:
-                LayersConfig._layer_name_list.append(name)
+        self.name = name
 
     def print_params(self, details=True, session=None):
         """Print all info of parameters in the network"""
@@ -435,9 +367,7 @@ def __str__(self):
         return "  Last layer is: %s (%s) %s" % (self.__class__.__name__, self.name, self.outputs.get_shape().as_list())
 
     def __getitem__(self, key):
-        set_name_reuse(True)
         net_new = Layer(self.inputs, name=self.name)
-        set_name_reuse(LayersConfig._name_reuse)  # set back
         net_new.outputs = self.outputs[key]
 
         net_new.all_layers = list(self.all_layers[:-1])
diff --git a/tensorlayer/layers/time_distribution.py b/tensorlayer/layers/time_distribution.py
@@ -63,9 +63,9 @@ def __init__(
         timestep = input_shape[1]
         x = tf.unstack(self.inputs, axis=1)
 
+        is_name_reuse = tf.get_variable_scope().reuse
         for i in range(0, timestep):
-            with tf.variable_scope(name, reuse=(LayersConfig._name_reuse if i == 0 else True)) as vs:
-                set_name_reuse((LayersConfig._name_reuse if i == 0 else True))
+            with tf.variable_scope(name, reuse=(is_name_reuse if i == 0 else True)) as vs:
                 net = layer_class(InputLayer(x[i], name=args['name'] + str(i)), **args)
                 x[i] = net.outputs
                 variables = tf.get_collection(TF_GRAPHKEYS_VARIABLES, scope=vs.name)
diff --git a/tests/test_mnist_simple.py b/tests/test_mnist_simple.py
@@ -15,10 +15,10 @@
 # define the network
 network = tl.layers.InputLayer(x, name='input')
 network = tl.layers.DropoutLayer(network, keep=0.8, name='drop1')
-network = tl.layers.DenseLayer(network, 800, tf.nn.relu, name='relu1')
-network = tl.layers.DropoutLayer(network, keep=0.5, name='drop2')
-network = tl.layers.DenseLayer(network, 800, tf.nn.relu, name='relu2')
-network = tl.layers.DropoutLayer(network, keep=0.5, name='drop3')
+network = tl.layers.DenseLayer(network, 100, tf.nn.relu, name='relu1')
+network = tl.layers.DropoutLayer(network, keep=0.8, name='drop2')
+network = tl.layers.DenseLayer(network, 100, tf.nn.relu, name='relu2')
+network = tl.layers.DropoutLayer(network, keep=0.8, name='drop3')
 # the softmax is implemented internally in tl.cost.cross_entropy(y, y_) to
 # speed up computation, so we use identity here.
 # see tf.nn.sparse_softmax_cross_entropy_with_logits()
@@ -44,7 +44,7 @@
 
 # train the network
 tl.utils.fit(
-    sess, network, train_op, cost, X_train, y_train, x, y_, acc=acc, batch_size=500, n_epoch=3, print_freq=1, X_val=X_val, y_val=y_val, eval_train=False)
+    sess, network, train_op, cost, X_train, y_train, x, y_, acc=acc, batch_size=500, n_epoch=1, print_freq=1, X_val=X_val, y_val=y_val, eval_train=False)
 
 # evaluation
 tl.utils.test(sess, network, acc, X_test, y_test, x, y_, batch_size=None, cost=cost)
diff --git a/tests/test_reuse_mlp.py b/tests/test_reuse_mlp.py
@@ -0,0 +1,40 @@
+import tensorflow as tf
+import tensorlayer as tl
+
+sess = tf.InteractiveSession()
+
+# prepare data
+X_train, y_train, X_val, y_val, X_test, y_test = \
+                                tl.files.load_mnist_dataset(shape=(-1,784))
+# define placeholder
+x = tf.placeholder(tf.float32, shape=[None, 784], name='x')
+y_ = tf.placeholder(tf.int64, shape=[None], name='y_')
+
+
+# define the network
+def mlp(x, is_train=True, reuse=False):
+    with tf.variable_scope("MLP", reuse=reuse):
+        tl.layers.set_name_reuse(reuse)  # print warning
+        network = tl.layers.InputLayer(x, name='input')
+        network = tl.layers.DropoutLayer(network, keep=0.8, is_fix=True, is_train=is_train, name='drop1')
+        network = tl.layers.DenseLayer(network, n_units=800, act=tf.nn.relu, name='relu1')
+        network = tl.layers.DropoutLayer(network, keep=0.5, is_fix=True, is_train=is_train, name='drop2')
+        network = tl.layers.DenseLayer(network, n_units=800, act=tf.nn.relu, name='relu2')
+        network = tl.layers.DropoutLayer(network, keep=0.5, is_fix=True, is_train=is_train, name='drop3')
+        network = tl.layers.DenseLayer(network, n_units=10, act=tf.identity, name='output')
+    return network
+
+
+# define inferences
+net_train = mlp(x, is_train=True, reuse=False)
+net_test = mlp(x, is_train=False, reuse=True)
+
+try:
+    is_except = False
+    net_test = mlp(x, is_train=False, reuse=False)
+except Exception as e:
+    is_except = True
+    print(e)
+
+if is_except == False:
+    raise Exception("it should not be success")
diff --git a/tests/test_time_distributed.py b/tests/test_time_distributed.py
@@ -0,0 +1,31 @@
+import tensorflow as tf
+from tensorlayer.layers import InputLayer, TimeDistributedLayer, DenseLayer
+
+sess = tf.InteractiveSession()
+batch_size = 32
+timestep = 20
+input_dim = 100
+
+## no reuse
+x = tf.placeholder(dtype=tf.float32, shape=[batch_size, timestep, input_dim], name="encode_seqs")
+net = InputLayer(x, name='input')
+net = TimeDistributedLayer(net, layer_class=DenseLayer, args={'n_units': 50, 'name': 'dense'}, name='time_dense')
+
+if net.outputs.get_shape().as_list() != [32, 20, 50]:
+    raise Exception("shape dont match")
+# ... (32, 20, 50)
+net.print_params(False)
+if net.count_params() != 5050:
+    raise Exception("params dont match")
+
+
+## reuse
+def model(x, is_train=True, reuse=False):
+    with tf.variable_scope("model", reuse=reuse):
+        net = InputLayer(x, name='input')
+        net = TimeDistributedLayer(net, layer_class=DenseLayer, args={'n_units': 50, 'name': 'dense'}, name='time_dense')
+    return net
+
+
+net_train = model(x, is_train=True, reuse=False)
+net_test = model(x, is_train=False, reuse=True)
