Remove tl.layers.initialize_global_variables(sess) (#931)

zsdonghao · web-flow · commit 94d20f71a3b9 · 2019-01-16T16:46:44.000+08:00
* update sampling layers

* upadte zoom

* fix bug zoom

* typo

* fix bug affine_transform_cv2 x and y

* fix bug crop when crop size equal to image size

* fix file docs typo

* fix bug instance norm

* fix docs

* update examples , init variables

* changelog
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -72,6 +72,7 @@ To release a new version, please update the changelog as followed:
 ### Added
 
 ### Changed
+- remove `tl.layers.initialize_global_variables(sess)` (PR #931)
 
 ### Dependencies Update
 - nltk>=3.3,<3.4 => nltk>=3.3,<3.5 (PR #892)
@@ -87,6 +88,7 @@ To release a new version, please update the changelog as followed:
 ### Security
 
 ### Contributors
+@zsdonghao: #931
 
 ## [1.11.1] - 2018-11-15
 
diff --git a/docs/modules/layers.rst b/docs/modules/layers.rst
@@ -39,7 +39,7 @@ All TensorLayer layers have a number of properties in common:
 
 All TensorLayer layers have a number of methods in common:
 
- - ``layer.print_params()`` : print network variable information in order (after ``tl.layers.initialize_global_variables(sess)``). alternatively, print all variables by ``tl.layers.print_all_variables()``.
+ - ``layer.print_params()`` : print network variable information in order (after ``sess.run(tf.global_variables_initializer())``). alternatively, print all variables by ``tl.layers.print_all_variables()``.
  - ``layer.print_layers()`` : print network layer information in order.
  - ``layer.count_params()`` : print the number of parameters in the network.
 
@@ -89,7 +89,7 @@ To count the number of parameters in a network, run ``network.count_params()``.
   train_op = tf.train.AdamOptimizer(learning_rate, beta1=0.9, beta2=0.999,
                               epsilon=1e-08, use_locking=False).minimize(cost, var_list = train_params)
 
-  tl.layers.initialize_global_variables(sess)
+  sess.run(tf.global_variables_initializer())
 
   network.print_params()
   network.print_layers()
diff --git a/examples/basic_tutorials/tutorial_cifar10_placeholder.py b/examples/basic_tutorials/tutorial_cifar10_placeholder.py
@@ -131,7 +131,7 @@ def distort_fn(x, is_train=False):
 train_op = tf.train.AdamOptimizer(learning_rate, beta1=0.9, beta2=0.999, epsilon=1e-08,
                                   use_locking=False).minimize(cost, var_list=train_params)
 
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 network.print_params(False)
 network.print_layers()
diff --git a/examples/basic_tutorials/tutorial_cifar10_tfrecord.py b/examples/basic_tutorials/tutorial_cifar10_tfrecord.py
@@ -277,7 +277,7 @@ def model_batch_norm(x_crop, y_, reuse, is_train):
     with tf.device('/gpu:0'):  # <-- remove it if you don't have GPU
         train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost)
 
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     if resume:
         print("Load existing model " + "!" * 10)
         saver = tf.train.Saver()
diff --git a/examples/basic_tutorials/tutorial_mlp_dropout1.py b/examples/basic_tutorials/tutorial_mlp_dropout1.py
@@ -37,7 +37,7 @@
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 # print network information
 network.print_params()
diff --git a/examples/basic_tutorials/tutorial_mlp_dropout2.py b/examples/basic_tutorials/tutorial_mlp_dropout2.py
@@ -46,7 +46,7 @@ def mlp(x, is_train=True, reuse=False):
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 n_epoch = 500
 batch_size = 500
diff --git a/examples/basic_tutorials/tutorial_mnist_autoencoder_cnn.py b/examples/basic_tutorials/tutorial_mnist_autoencoder_cnn.py
@@ -76,7 +76,7 @@ def main_test_layers(model='relu'):
     print_freq = 5
     train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost)
 
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     net.print_params()
     net.print_layers()
@@ -179,7 +179,7 @@ def main_test_denoise_AE(model='relu'):
         recon_layer1 = tl.layers.ReconLayer(net, x_recon=x, n_units=784, act=tf.nn.sigmoid, name='recon_layer1')
 
     # ready to train
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     # print all params
     print("All net Params")
@@ -253,7 +253,7 @@ def main_test_stacked_denoise_AE(model='relu'):
     train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost, var_list=train_params)
 
     # Initialize all variables including weights, biases and the variables in train_op
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     # Pre-train
     print("\nAll net Params before pre-train")
@@ -417,7 +417,7 @@ def main_test_cnn_layer():
     train_params = net.all_params
     train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost, var_list=train_params)
 
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     net.print_params()
     net.print_layers()
 
diff --git a/examples/basic_tutorials/tutorial_mnist_float16.py b/examples/basic_tutorials/tutorial_mnist_float16.py
@@ -64,7 +64,7 @@ def model(x, is_train=True, reuse=False):
                                   use_locking=False).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 # train the network
 n_epoch = 500
diff --git a/examples/basic_tutorials/tutorial_mnist_simple.py b/examples/basic_tutorials/tutorial_mnist_simple.py
@@ -39,7 +39,7 @@
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 # print network information
 network.print_params()
diff --git a/examples/database/task_script.py b/examples/database/task_script.py
@@ -53,7 +53,7 @@ def mlp(x, is_train=True, reuse=False):
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 # train the network
 tl.utils.fit(
diff --git a/examples/keras_tfslim/tutorial_keras.py b/examples/keras_tfslim/tutorial_keras.py
@@ -47,7 +47,7 @@ def keras_block(x):
 train_params = network.all_params
 train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost, var_list=train_params)
 
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 for epoch in range(n_epoch):
     start_time = time.time()
diff --git a/examples/keras_tfslim/tutorial_tfslim.py b/examples/keras_tfslim/tutorial_tfslim.py
@@ -48,7 +48,7 @@ def slim_block(x):
 train_params = network.all_params
 train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost, var_list=train_params)
 
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 for epoch in range(n_epoch):
     start_time = time.time()
diff --git a/examples/pretrained_cnn/tutorial_inceptionV3_tfslim.py b/examples/pretrained_cnn/tutorial_inceptionV3_tfslim.py
@@ -94,7 +94,7 @@ def print_prob(prob):
 #                                     )
 # sess = tf.InteractiveSession()
 # # sess.run(tf.global_variables_initializer())
-# tl.layers.initialize_global_variables(sess)
+# sess.run(tf.global_variables_initializer())
 # network.print_params()
 
 ## InceptionV3 / All TF-Slim nets can be merged into TensorLayer
diff --git a/examples/pretrained_cnn/tutorial_mobilenet.py b/examples/pretrained_cnn/tutorial_mobilenet.py
@@ -101,7 +101,7 @@ def mobilenet(x, is_train=True, reuse=False):
 n.print_params(False)
 
 sess = tf.InteractiveSession()
-# tl.layers.initialize_global_variables(sess)
+# sess.run(tf.global_variables_initializer())
 
 if not os.path.isfile(MODEL_PATH):
     raise Exception("Please download mobilenet.npz from : https://github.com/tensorlayer/pretrained-models")
diff --git a/examples/pretrained_cnn/tutorial_vgg16.py b/examples/pretrained_cnn/tutorial_vgg16.py
@@ -122,7 +122,7 @@ def fc_layers(net):
 # correct_prediction = tf.equal(tf.cast(tf.argmax(y, 1), tf.float32), tf.cast(y_, tf.float32))
 # acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
 
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 net.print_params()
 net.print_layers()
 
diff --git a/examples/pretrained_cnn/tutorial_vgg19.py b/examples/pretrained_cnn/tutorial_vgg19.py
@@ -212,7 +212,7 @@ def Vgg19_simple_api(rgb):
 net = Vgg19_simple_api(x)
 y = net.outputs
 probs = tf.nn.softmax(y, name="prob")
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 # You need to download the pre-trained model - VGG19 NPY
 if not os.path.isfile(MODEL_PATH):
diff --git a/examples/quantized_net/tutorial_binarynet_cifar10_tfrecord.py b/examples/quantized_net/tutorial_binarynet_cifar10_tfrecord.py
@@ -211,7 +211,7 @@ def model(x_crop, y_, reuse):
     with tf.device('/gpu:0'):  # <-- remove it if you don't have GPU
         train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost)
 
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     if resume:
         print("Load existing model " + "!" * 10)
         saver = tf.train.Saver()
diff --git a/examples/quantized_net/tutorial_binarynet_mnist_cnn.py b/examples/quantized_net/tutorial_binarynet_mnist_cnn.py
@@ -65,7 +65,7 @@ def model(x, is_train=True, reuse=False):
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 net_train.print_params()
 net_train.print_layers()
diff --git a/examples/quantized_net/tutorial_dorefanet_cifar10_tfrecord.py b/examples/quantized_net/tutorial_dorefanet_cifar10_tfrecord.py
@@ -207,7 +207,7 @@ def model(x_crop, y_, reuse):
     with tf.device('/gpu:0'):  # <-- remove it if you don't have GPU
         train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost)
 
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     if resume:
         print("Load existing model " + "!" * 10)
         saver = tf.train.Saver()
diff --git a/examples/quantized_net/tutorial_dorefanet_mnist_cnn.py b/examples/quantized_net/tutorial_dorefanet_mnist_cnn.py
@@ -65,7 +65,7 @@ def model(x, is_train=True, reuse=False):
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 net_train.print_params()
 net_train.print_layers()
diff --git a/examples/quantized_net/tutorial_ternaryweight_cifar10_tfrecord.py b/examples/quantized_net/tutorial_ternaryweight_cifar10_tfrecord.py
@@ -206,7 +206,7 @@ def model(x_crop, y_, reuse):
     with tf.device('/gpu:0'):  # <-- remove it if you don't have GPU
         train_op = tf.train.AdamOptimizer(learning_rate).minimize(cost)
 
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     if resume:
         print("Load existing model " + "!" * 10)
         saver = tf.train.Saver()
diff --git a/examples/quantized_net/tutorial_ternaryweight_mnist_cnn.py b/examples/quantized_net/tutorial_ternaryweight_mnist_cnn.py
@@ -65,7 +65,7 @@ def model(x, is_train=True, reuse=False):
 train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost, var_list=train_params)
 
 # initialize all variables in the session
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 net_train.print_params()
 net_train.print_layers()
diff --git a/examples/reinforcement_learning/tutorial_atari_pong.py b/examples/reinforcement_learning/tutorial_atari_pong.py
@@ -77,7 +77,7 @@ def prepro(I):
 train_op = tf.train.RMSPropOptimizer(learning_rate, decay_rate).minimize(loss)
 
 with tf.Session() as sess:
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     # if resume:
     #     load_params = tl.files.load_npz(name=model_file_name+'.npz')
     #     tl.files.assign_params(sess, load_params, network)
diff --git a/examples/reinforcement_learning/tutorial_bipedalwalker_a3c_continuous_action.py b/examples/reinforcement_learning/tutorial_bipedalwalker_a3c_continuous_action.py
@@ -263,7 +263,7 @@ def work(self):
             workers.append(Worker(i_name, GLOBAL_AC))
 
     COORD = tf.train.Coordinator()
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     # start TF threading
     worker_threads = []
@@ -278,7 +278,7 @@ def work(self):
     # ============================= EVALUATION =============================
     # env = gym.make(GAME)
     # GLOBAL_AC = ACNet(GLOBAL_NET_SCOPE)
-    # tl.layers.initialize_global_variables(sess)
+    # sess.run(tf.global_variables_initializer())
     # GLOBAL_AC.load_ckpt()
     # while True:
     #     s = env.reset()
diff --git a/examples/reinforcement_learning/tutorial_cartpole_ac.py b/examples/reinforcement_learning/tutorial_cartpole_ac.py
@@ -151,7 +151,7 @@ def learn(self, s, r, s_):
 # we need a good teacher, so the teacher should learn faster than the actor
 critic = Critic(sess, n_features=N_F, lr=LR_C)
 
-tl.layers.initialize_global_variables(sess)
+sess.run(tf.global_variables_initializer())
 
 if OUTPUT_GRAPH:
     tf.summary.FileWriter("logs/", sess.graph)
diff --git a/examples/reinforcement_learning/tutorial_frozenlake_dqn.py b/examples/reinforcement_learning/tutorial_frozenlake_dqn.py
@@ -68,7 +68,7 @@ def to_one_hot(i, n_classes=None):
 e = 0.1  # e-Greedy Exploration, the larger the more random
 num_episodes = 10000
 with tf.Session() as sess:
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     for i in range(num_episodes):
         ## Reset environment and get first new observation
         episode_time = time.time()
diff --git a/examples/text_classification/tutorial_imdb_fasttext.py b/examples/text_classification/tutorial_imdb_fasttext.py
@@ -129,7 +129,7 @@ def train_test_and_save_model():
     )
 
     with tf.Session() as sess:
-        tl.layers.initialize_global_variables(sess)
+        sess.run(tf.global_variables_initializer())
 
         for epoch in range(N_EPOCH):
             start_time = time.time()
diff --git a/examples/text_generation/tutorial_generate_text.py b/examples/text_generation/tutorial_generate_text.py
@@ -157,7 +157,7 @@ def main_restore_embedding_layer():
     emb_net = tl.layers.EmbeddingInputlayer(x, vocabulary_size, embedding_size, name='emb')
 
     # sess.run(tf.global_variables_initializer())
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     tl.files.assign_params(sess, [load_params[0]], emb_net)
 
@@ -283,7 +283,7 @@ def loss_fn(outputs, targets, batch_size, sequence_length):
     train_op = optimizer.apply_gradients(zip(grads, tvars))
 
     # ===== Training
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     print("\nStart learning a model to generate text")
     for i in range(max_max_epoch):
diff --git a/examples/text_ptb/tutorial_ptb_lstm.py b/examples/text_ptb/tutorial_ptb_lstm.py
@@ -246,7 +246,7 @@ def inference(x, is_training, num_steps, reuse=None):
     net_test, lstm1_test, lstm2_test = inference(input_data_test, is_training=False, num_steps=1, reuse=True)
 
     # sess.run(tf.global_variables_initializer())
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     def loss_fn(outputs, targets):  # , batch_size, num_steps):
         # See tl.cost.cross_entropy_seq()
@@ -280,8 +280,7 @@ def loss_fn(outputs, targets):  # , batch_size, num_steps):
     optimizer = tf.train.GradientDescentOptimizer(lr)
     train_op = optimizer.apply_gradients(zip(grads, tvars))
 
-    # sess.run(tf.global_variables_initializer())
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     net.print_params()
     net.print_layers()
diff --git a/examples/text_ptb/tutorial_ptb_lstm_state_is_tuple.py b/examples/text_ptb/tutorial_ptb_lstm_state_is_tuple.py
@@ -252,7 +252,7 @@ def inference(x, is_training, num_steps, reuse=None):
     net_test, lstm1_test, lstm2_test = inference(input_data_test, is_training=False, num_steps=1, reuse=True)
 
     # sess.run(tf.global_variables_initializer())
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     def loss_fn(outputs, targets, batch_size):
         # See tl.cost.cross_entropy_seq()
@@ -287,7 +287,7 @@ def loss_fn(outputs, targets, batch_size):
     train_op = optimizer.apply_gradients(zip(grads, tvars))
 
     # sess.run(tf.global_variables_initializer())
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
 
     net.print_params()
     net.print_layers()
diff --git a/examples/text_word_embedding/tutorial_word2vec_basic.py b/examples/text_word_embedding/tutorial_word2vec_basic.py
@@ -218,9 +218,7 @@ def main_word2vec_basic():
     # transpose_b=True, normalized_embeddings is transposed before multiplication.
 
     # Step 5: Start training.
-    print()
-
-    tl.layers.initialize_global_variables(sess)
+    sess.run(tf.global_variables_initializer())
     if resume:
         print("Load existing model" + "!" * 10)
         # Load from ckpt or npz file
