All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/)

- API:

- Add `tl.model.vgg19` (PR #698)

- Add `tl.logging.contrib.hyperdash` (PR #739)

- `tutorial_mnist_distributed_trainer.py` and `tutorial_cifar10_distributed_trainer.py` are added to explain the uses of Distributed Trainer (PR #700)

- add `tutorial_quanconv_cifar10.py` and `tutorial_quanconv_mnist.py` (PR #735)

- add `tutorial_work_with_onnx.py`(PR #775)

- function minibatches changed to avoid wasting samples.(PR #762)

- all the input scale in both vgg16 and vgg19 has been changed the input scale from [0,255] to [0,1](PR #710)

- Dockerfiles merged and refactored into one file (PR #747)

- LazyImports move to the most **top level** imports as possible (PR #739)

- some new test functions have been added in `test_layers_convolution.py`, `test_layers_normalization.py`, `test_layers_core.py` (PR #735)

- documentation now uses mock imports reducing the number of dependencies to compile the documentation (PR #785)

- `tl.logging.warn` has been deprecated in favor of `tl.logging.warning` (PR #739)

- `conv_layers()` has been removed in both vgg16 and vgg19(PR #710)

- import error caused by matplotlib on OSX (PR #705)

- missing import in tl.prepro (PR #712)

- Dockerfiles import error fixed - issue #733 (PR #747)

- tensorflow>=1.8,<1.9 => tensorflow>=1.9,<1.10 (PR #739)

- tensorflow-gpu>=1.8,<1.9 => tensorflow-gpu>=1.9,<1.10 (PR #739)

- pymongo>=3.6,<3.7 => pymongo>=3.7,<3.8 (PR #750)

- numpy>=1.14,<1.15 => numpy>=1.15,<1.16 (PR #754)

- @DEKHTIARJonathan: #739 #747 #750 #754

- @lgarithm: #705 #700

- @OwenLiuzZ: #698 #710 #775 #776

- API:

- `tl.alphas` and `tl.alphas_like` added following the tf.ones/zeros and tf.zeros_like/ones_like (PR #580)

- `tl.lazy_imports.LazyImport` to import heavy libraries only when necessary (PR #667)

```

- Tensorflow CPU & GPU dependencies moved to separated requirement files in order to allow PyUP.io to parse them (PR #573)

- The document of LambdaLayer for linking it with ElementwiseLambdaLayer (PR #587)

- RTD links point to stable documentation instead of latest used for development (PR #633)

- Complete Documentation Refactoring and Reorganization (namely Layer APIs) (PR #691)

- `tl.layers.TimeDistributedLayer` argurment `args` is deprecated in favor of `layer_args` (PR #667)

- `tl.act.leaky_relu` have been deprecated in favor of `tf.nn.leaky_relu` (PR #686)

- `assert()` calls remove and replaced by `raise AssertionError()` (PR #667)

- `tl.identity` is removed, not used anymore and deprecated for a long time (PR #667)

- All Code specific to `TF.__version__ < "1.6"` have been removed (PR #675)

- Issue #498 - Deprecation Warning Fix in `tl.layers.RNNLayer` with `inspect` (PR #574)

- Issue #498 - Deprecation Warning Fix in `tl.files` with truth value of an empty array is ambiguous (PR #575)

- Issue #565 related to `tl.utils.predict` fixed - `np.hstack` problem in which the results for multiple batches are stacked along `axis=1` (PR #566)

- All tutorials tested and errors have been fixed (PR #635)

- Update pytest from 3.5.1 to 3.6.0 (PR #647)

- Update progressbar2 from 3.37.1 to 3.38.0 (PR #651)

- Update scikit-image from 0.13.1 to 0.14.0 (PR #656)

- Update keras from 2.1.6 to 2.2.0 (PR #684)

- Update requests from 2.18.4 to 2.19.0 (PR #695)

- @lgarithm: #563

- @DEKHTIARJonathan: #573 #574 #575 #580 #633 #635 #636 #639 #644 #645 #648 #657 #667 #658 #659 #660 #661 #666 #667 #672 #675 #683 #686 #687 #690 #691 #692 #703

- @2wins: #560 #566 #662

- Github Templates added (by @DEKHTIARJonathan)

- New issues Template

- New PR Template

- `test_layers_core` has been added to ensure that `LayersConfig` is abstract.

- All Tests Refactored - Now using unittests and runned with PyTest (by @DEKHTIARJonathan)

- Documentation updated (by @zsdonghao)

- Package Setup Refactored (by @DEKHTIARJonathan)

- YAPF coding style improved and enforced (by @DEKHTIARJonathan)

- Backward Compatibility Restored with deprecation warnings (by @DEKHTIARJonathan)

- Tensorflow Deprecation Fix (Issue #498):

- AverageEmbeddingInputlayer (by @zsdonghao)

- Pooling Layer Issue #557 fixed (by @zsdonghao)

- scipy>=1.0,<1.1 => scipy>=1.1,<1.2

@zsdonghao @luomai @DEKHTIARJonathan

[Unreleased]: https://github.com/tensorlayer/tensorlayer/compare/1.9.0...master

[1.9.0]: https://github.com/tensorlayer/tensorlayer/compare/1.9.0...1.8.5

[1.8.5]: https://github.com/tensorlayer/tensorlayer/compare/1.8.4...1.8.5

'sklearn',

'tensorflow',

'tqdm',

]

import tensorflow as tf

import tensorlayer as tl

tl.layers.initialize_global_variables(sess)

test_accuracy = tl.utils.test(sess, net_test, acc_test, X_test, y_test, x, y_, batch_size=None, cost=cost_test)

def prepro(I):

I = I[35:195]

I = I[::2, ::2, 0]

I[I == 144] = 0

start_time = time.time()

game_number = 0

while True:

cur_x = prepro(observation)

x = cur_x - prev_x if prev_x is not None else np.zeros(D)

prev_x = None

if reward != 0:

start_time = time.time()

game_number += 1

model_file_name = "./model_cifar10_tfrecord.ckpt"

resume = False # load model, resume from previous checkpoint?

X_train, y_train, X_test, y_test = tl.files.load_cifar10_dataset(shape=(-1, 32, 32, 3), plotable=False)

print('X_train.shape', X_train.shape) # (50000, 32, 32, 3)

def data_to_tfrecord(images, labels, filename):

if os.path.isfile(filename):

print("%s exists" % filename)

return

writer = tf.python_io.TFRecordWriter(filename)

for index, img in enumerate(images):

img_raw = img.tobytes()

# tl.visualize.frame(np.asarray(img, dtype=np.uint8), second=1, saveable=False, name='frame', fig_idx=1236)

label = int(labels[index])

# print(label)

# image = Image.frombytes('RGB', (32, 32), img_raw)

# image = np.fromstring(img_raw, np.float32)

# image = image.reshape([32, 32, 3])

def read_and_decode(filename, is_train=None):

filename_queue = tf.train.string_input_producer([filename])

reader = tf.TFRecordReader()

_, serialized_example = reader.read(filename_queue)

return img, label

data_to_tfrecord(images=X_train, labels=y_train, filename="train.cifar10")

data_to_tfrecord(images=X_test, labels=y_test, filename="test.cifar10")

)

def model(x_crop, y_, reuse):

with tf.variable_scope("model", reuse=reuse):

net = tl.layers.InputLayer(x_crop, name='input')

net = tl.layers.Conv2d(net, 64, (5, 5), (1, 1), act=tf.nn.relu, padding='SAME', name='cnn1')

return net, cost, acc

# x_crop = tf.placeholder(tf.float32, shape=[batch_size, 24, 24, 3])

# y_ = tf.placeholder(tf.int32, shape=[batch_size,])

# cost, acc, network = model(x_crop, y_, None)

network, cost, acc, = model(x_train_batch, y_train_batch, False)

_, cost_test, acc_test = model(x_test_batch, y_test_batch, True)

n_epoch = 50000

learning_rate = 0.0001

print_freq = 1

start_time = time.time()

train_loss, train_acc, n_batch = 0, 0, 0

for s in range(n_step_epoch):

# val, l = sess.run([x_train_batch, y_train_batch])

# tl.visualize.images2d(val, second=3, saveable=False, name='batch', dtype=np.uint8, fig_idx=2020121)

# err, ac, _ = sess.run([cost, acc, train_op], feed_dict={x_crop: val, y_: l})

def model_batch_norm(x, y_, reuse, is_train):

W_init = tf.truncated_normal_initializer(stddev=5e-2)

W_init2 = tf.truncated_normal_initializer(stddev=0.04)

b_init2 = tf.constant_initializer(value=0.1)

x = tf.placeholder(dtype=tf.float32, shape=[None, 24, 24, 3], name='x')

y_ = tf.placeholder(dtype=tf.int64, shape=[None], name='y_')

network, cost, _ = model_batch_norm(x, y_, False, is_train=True)

_, cost_test, acc = model_batch_norm(x, y_, True, is_train=False)