use yapf

Author: Laicheng0830

examples/quantized_net/tutorial_binarynet_cifar10.py renamed to examples/quantized_net/tutorial_binarynet_cifar10_tfrecord.py

@@ -45,9 +45,9 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (BinaryConv2d, BinaryDense, Conv2d, Dense,
-                                Flatten, Input, LocalResponseNorm, MaxPool2d,
-                                Sign)
+from tensorlayer.layers import (
+    BinaryConv2d, BinaryDense, Conv2d, Dense, Flatten, Input, LocalResponseNorm, MaxPool2d, Sign
+)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
@@ -78,6 +78,7 @@ def binary_model(input_shape, n_classes):
     net = Model(inputs=in_net, outputs=net, name='binarynet')
     return net
 
+
 # training settings
 net = binary_model([None, 24, 24, 3], n_classes=10)
 batch_size = 128
@@ -92,6 +93,7 @@ def binary_model(input_shape, n_classes):
 optimizer = tf.optimizers.Adam(learning_rate)
 cost = tl.cost.cross_entropy
 
+
 def generator_train():
     inputs = X_train
     targets = y_train
@@ -137,7 +139,6 @@ def _map_fn_test(img, target):
     return img, target
 
 
-
 def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None):
     with tf.GradientTape() as tape:
         y_pred = network(X_batch)
@@ -150,9 +151,11 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 # dataset API and augmentation
 train_ds = tf.data.Dataset.from_generator(
     generator_train, output_types=(tf.float32, tf.int32)

examples/quantized_net/tutorial_binarynet_mnist_cnn.py

@@ -6,8 +6,7 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (BatchNorm, BinaryConv2d, BinaryDense, Flatten,
-                                Input, MaxPool2d, Sign)
+from tensorlayer.layers import (BatchNorm, BinaryConv2d, BinaryDense, Flatten, Input, MaxPool2d, Sign)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
@@ -16,10 +15,11 @@
 
 batch_size = 128
 
+
 def model(inputs_shape, n_class=10):
     # In BNN, all the layers inputs are binary, with the exception of the first layer.
     # ref: https://github.com/itayhubara/BinaryNet.tf/blob/master/models/BNN_cifar10.py
-    net_in = Input(inputs_shape,name='input')
+    net_in = Input(inputs_shape, name='input')
     net = BinaryConv2d(32, (5, 5), (1, 1), padding='SAME', b_init=None, name='bcnn1')(net_in)
     net = MaxPool2d((2, 2), (2, 2), padding='SAME', name='pool1')(net)
     net = BatchNorm(act=tl.act.htanh, name='bn1')(net)
@@ -53,17 +53,18 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 n_epoch = 200
 print_freq = 5
 
 net = model([None, 28, 28, 1])
 train_op = tf.optimizers.Adam(learning_rate=0.0001)
 cost = tl.cost.cross_entropy
 
-
 for epoch in range(n_epoch):
     start_time = time.time()
     train_loss, train_acc, n_batch = 0, 0, 0
@@ -93,7 +94,6 @@ def accuracy(_logits, y_batch):
         print("   val loss: {}".format(val_loss / val_batch))
         print("   val acc:  {}".format(val_acc / val_batch))
 
-
 net.test()
 test_loss, test_acc, n_test_batch = 0, 0, 0
 for X_test_a, y_test_a in tl.iterate.minibatches(X_test, y_test, batch_size, shuffle=True):

examples/quantized_net/tutorial_dorefanet_cifar10.py renamed to examples/quantized_net/tutorial_dorefanet_cifar10_tfrecord.py

@@ -45,8 +45,7 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (Conv2d, Dense, DorefaConv2d, DorefaDense,
-                                Flatten, Input, LocalResponseNorm, MaxPool2d)
+from tensorlayer.layers import (Conv2d, Dense, DorefaConv2d, DorefaDense, Flatten, Input, LocalResponseNorm, MaxPool2d)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
@@ -72,6 +71,7 @@ def dorefanet_model(input_shape, n_classes):
     net = Model(inputs=in_net, outputs=net, name='dorefanet')
     return net
 
+
 # training settings
 net = dorefanet_model([None, 24, 24, 3], n_classes=10)
 batch_size = 128
@@ -86,6 +86,7 @@ def dorefanet_model(input_shape, n_classes):
 # optimizer = tf.optimizers.SGD(learning_rate)
 cost = tl.cost.cross_entropy
 
+
 def generator_train():
     inputs = X_train
     targets = y_train
@@ -131,7 +132,6 @@ def _map_fn_test(img, target):
     return img, target
 
 
-
 def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None):
     with tf.GradientTape() as tape:
         y_pred = network(X_batch)
@@ -144,9 +144,11 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 # dataset API and augmentation
 train_ds = tf.data.Dataset.from_generator(
     generator_train, output_types=(tf.float32, tf.int32)

examples/quantized_net/tutorial_dorefanet_mnist_cnn.py

@@ -6,8 +6,7 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (BatchNorm, Dense, DorefaConv2d, DorefaDense,
-                                Flatten, Input, MaxPool2d)
+from tensorlayer.layers import (BatchNorm, Dense, DorefaConv2d, DorefaDense, Flatten, Input, MaxPool2d)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
@@ -16,6 +15,7 @@
 
 batch_size = 128
 
+
 def model(inputs_shape, n_class=10):
     in_net = Input(inputs_shape, name='input')
     net = DorefaConv2d(1, 3, 32, (5, 5), (1, 1), padding='SAME', b_init=None, name='bcnn1')(in_net)
@@ -48,17 +48,18 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 n_epoch = 200
 print_freq = 5
 
 net = model([None, 28, 28, 1])
 train_op = tf.optimizers.Adam(learning_rate=0.0001)
 cost = tl.cost.cross_entropy
 
-
 for epoch in range(n_epoch):
     start_time = time.time()
     train_loss, train_acc, n_batch = 0, 0, 0
@@ -88,7 +89,6 @@ def accuracy(_logits, y_batch):
         print("   val loss: {}".format(val_loss / val_batch))
         print("   val acc:  {}".format(val_acc / val_batch))
 
-
 net.test()
 test_loss, test_acc, n_test_batch = 0, 0, 0
 for X_test_a, y_test_a in tl.iterate.minibatches(X_test, y_test, batch_size, shuffle=True):

examples/quantized_net/tutorial_quanconv_cifar10.py

@@ -44,8 +44,7 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (Dense, Flatten, Input, MaxPool2d,
-                                QuanConv2dWithBN, QuanDense)
+from tensorlayer.layers import (Dense, Flatten, Input, MaxPool2d, QuanConv2dWithBN, QuanDense)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
@@ -68,6 +67,7 @@ def model(input_shape, n_classes, bitW, bitA):
     net = Model(inputs=in_net, outputs=net, name='dorefanet')
     return net
 
+
 # training settings
 bitW = 8
 bitA = 8
@@ -83,6 +83,7 @@ def model(input_shape, n_classes, bitW, bitA):
 optimizer = tf.optimizers.Adam(learning_rate)
 cost = tl.cost.cross_entropy
 
+
 def generator_train():
     inputs = X_train
     targets = y_train
@@ -128,7 +129,6 @@ def _map_fn_test(img, target):
     return img, target
 
 
-
 def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None):
     with tf.GradientTape() as tape:
         y_pred = network(X_batch)
@@ -141,9 +141,11 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 # dataset API and augmentation
 train_ds = tf.data.Dataset.from_generator(
     generator_train, output_types=(tf.float32, tf.int32)

examples/quantized_net/tutorial_quanconv_mnist.py

@@ -6,27 +6,31 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (Dense, Dropout, Flatten, Input, MaxPool2d,
-                                QuanConv2d, QuanConv2dWithBN, QuanDense,
-                                QuanDenseLayerWithBN)
+from tensorlayer.layers import (
+    Dense, Dropout, Flatten, Input, MaxPool2d, QuanConv2d, QuanConv2dWithBN, QuanDense, QuanDenseLayerWithBN
+)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
 
-
 X_train, y_train, X_val, y_val, X_test, y_test = tl.files.load_mnist_dataset(shape=(-1, 28, 28, 1))
 # X_train, y_train, X_test, y_test = tl.files.load_cropped_svhn(include_extra=False)
 
 batch_size = 128
 
+
 def model(inputs_shape, n_class=10):
     net_in = Input(inputs_shape, name="input")
 
-    net = QuanConv2dWithBN(n_filter=32, filter_size=(5,5),strides=(1,1),padding='SAME',act=tl.nn.relu, name='qconvbn1')(net_in)
-    net = MaxPool2d(filter_size=(2,2),strides=(2,2),padding='SAME',name='pool1')(net)
+    net = QuanConv2dWithBN(
+        n_filter=32, filter_size=(5, 5), strides=(1, 1), padding='SAME', act=tl.nn.relu, name='qconvbn1'
+    )(net_in)
+    net = MaxPool2d(filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool1')(net)
 
-    net = QuanConv2dWithBN(n_filter=64, filter_size=(5,5),strides=(1,1),padding='SAME',act=tl.nn.relu, name='qconvbn2')(net)
-    net = MaxPool2d(filter_size=(2,2),strides=(2,2),padding='SAME',name='pool2')(net)
+    net = QuanConv2dWithBN(
+        n_filter=64, filter_size=(5, 5), strides=(1, 1), padding='SAME', act=tl.nn.relu, name='qconvbn2'
+    )(net)
+    net = MaxPool2d(filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool2')(net)
 
     net = Flatten(name='ft')(net)
 
@@ -42,6 +46,7 @@ def model(inputs_shape, n_class=10):
     net = Model(inputs=net_in, outputs=net, name='quan')
     return net
 
+
 def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None):
     with tf.GradientTape() as tape:
         y_pred = network(X_batch)
@@ -54,9 +59,11 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 n_epoch = 200
 print_freq = 1
 
@@ -97,7 +104,6 @@ def accuracy(_logits, y_batch):
         print("   val loss: {}".format(val_loss / n_eval))
         print("   val acc:  {}".format(val_acc / n_eval))
 
-
 # net.eval()
 test_loss, test_acc, n_test_batch = 0, 0, 0
 for X_test_a, y_test_a in tl.iterate.minibatches(X_test, y_test, batch_size, shuffle=True):

examples/quantized_net/tutorial_ternaryweight_cifar10.py renamed to examples/quantized_net/tutorial_ternaryweight_cifar10_tfrecord.py

@@ -44,9 +44,9 @@
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
-from tensorlayer.layers import (Conv2d, Dense, Flatten, Input,
-                                LocalResponseNorm, MaxPool2d, TernaryConv2d,
-                                TernaryDense)
+from tensorlayer.layers import (
+    Conv2d, Dense, Flatten, Input, LocalResponseNorm, MaxPool2d, TernaryConv2d, TernaryDense
+)
 from tensorlayer.models import Model
 
 tl.logging.set_verbosity(tl.logging.DEBUG)
@@ -76,6 +76,7 @@ def model(input_shape, n_classes):
     net = Model(inputs=in_net, outputs=net, name='dorefanet')
     return net
 
+
 # training settings
 bitW = 8
 bitA = 8
@@ -91,6 +92,7 @@ def model(input_shape, n_classes):
 optimizer = tf.optimizers.Adam(learning_rate)
 cost = tl.cost.cross_entropy
 
+
 def generator_train():
     inputs = X_train
     targets = y_train
@@ -136,7 +138,6 @@ def _map_fn_test(img, target):
     return img, target
 
 
-
 def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None):
     with tf.GradientTape() as tape:
         y_pred = network(X_batch)
@@ -149,9 +150,11 @@ def _train_step(network, X_batch, y_batch, cost, train_op=tf.optimizers.Adam(lea
     else:
         return _loss, None
 
+
 def accuracy(_logits, y_batch):
     return np.mean(np.equal(np.argmax(_logits, 1), y_batch))
 
+
 # dataset API and augmentation
 train_ds = tf.data.Dataset.from_generator(
     generator_train, output_types=(tf.float32, tf.int32)