Maintenance TensorLayer 1.7.1

This is an incomplete list among the many features added.

• New Example

  • Float 16 half-precision model, see tutorial_mnist_float16.py by @zhonghao
  • Data augmentation for object detection by @zsdonghao

• New Support

  • tl.prepro.obj_box_coord_upleft_butright_to_centroid by @zsdonghao

• New Update

  • Fix bug of DeformableConv2dLayer by @vyokky
  • BatchNormLayer supports float 16 c99a6a2 by @zsdonghao

• Documentation by @vyokky @zsdonghao

Release 1.7.0

This is an incomplete list among the many features added.

• New Support

  • ConvLSTMLayer for Convolutional LSTM by @vyokky
  • DeformableConv2dLayer for Deformable Convolution by @vyokky
  • Many APIs for data augmentation for object detection, see tl.prepro by @zsdonghao
  • Download VOC dataset in one line of code, see load_voc_dataset by @zsdonghao
  • Download CelebA dataset in one line of code, see load_celebA_dataset by @zsdonghao

• New Update

  • Fix bug of tl.utils.predict by @xionghhcs 7d5e21b

• Documentation by @zsdonghao

Release 1.6.7

This is an incomplete list among the many features added.

• New Support

  • tl.prepro.shear2 for shearing of x and y directions.
  • tl.cost.absolute_difference_error for 2d, 2d and 4d tensor.
  • DepthwiseConv2d for separable/depthwise convolution, see tf.nn.depthwise_conv2d.

• New Update

  • SubpixelConv2d with tf.depth_to_space by @ShieLian
  • tl.prepro.illumination support random mode by @zsdonghao
  • tl.prepro.projective_transform_by_points for projective transformation by given points

• EN/CN Documentation by @zsdonghao

• ACM MM [slides]

Release 1.6.6

First of all, we win the Best Open Source Software Award @ACM MM 2017.

This is an incomplete list among the many features added.

• New Support

  • tl.prepro.illumination for changing image illumination by @yzwxx
  • tl.act.swish for Swish activation by @zsdonghao

• New Update

  • new scope name for tl.act.lrelu by @zsdonghao
  • more description for RL examples by @zsdonghao
  • AverageEmbeddingInputlayer for py3 by @zsdonghao
  • fixed bug of Seq2Seq : not saving parameters of encoder part by @zsdonghao

• Documentation by @zsdonghao @luomai

Release 1.6.5

This is an incomplete list among the many features added.

• New Example

  • FastText Sentence Classification (IMDB), see tutorial_imdb_fasttext.py by @tomtung
  • Variational Autoencoder (VAE) for (MNIST) by @BUPTLdy.

• New Support

  • AverageEmbeddingInputLayer by @tomtung

• New Update

  • tl.files.load_cyclegan_dataset for 2d grey scale to RGB.
  • tl.prepro.threading_data for returning list instead of numpy array if possible.
  • all examples to support TensorFlow 1.3 by @zsdonghao, see (6430b24)

• Documentation by @zsdonghao @luomai

Maintenance Release 1.6.4

This is an incomplete list among the many features added.

• New example

  • Improved CycleGAN with resize-convolution for unsupervised image-to-image translation by @luoxier

• New support

  • LayerNormLayer see tf.contrib.layers.layer_norm by @zsdonghao

• New update

  • tl.cost.mean_squared_error (0ede665) for dim=3 i.e. [batch_size, w, h] by @zsdonghao
  • Fix wrong default parameter (b22d37b) for tl.prepro.imresize by Yu Fang (@squarefk)
  • More pyplot (8b2c382) in tl.vis to the functions by @zsdonghao

• Documentation by @luomai @zsdonghao

Release 1.6.3

This is an incomplete list among the many features added.

• New example

  • Chatbot in 200 lines of code for Seq2Seq

• New Support

  • tl.nlp.moses_multi_bleu for BLEU by @zsdonghao
  • tl.prepro.remove_pad_sequences, tl.prepro.sequences_add_end_id and tl.prepro.sequences_add_end_id_after_pad for sequence preprocessing by @zsdonghao

• New Update

  • tl.files.load_and_assign_npz_dict uses GLOBAL_VARIABLES instead of TRAINABLE_VARIABLES by @Lucklyric
  • Allowed name to be reused in TimeDistributedLayer by Dan Busbridge
  • [tl.prepro.pad_sequences] returns list instead of numpy by @zsdonghao

• Documentation @zsdonghao

Release 1.6.2

This is an incomplete list among the many features added.

• New Support

  • Region of interest pooling layer for Object Detection by @zsdonghao
  • Stack and UnStack layer by @zsdonghao
  • Image Erosion by @zsdonghao
  • tl.prepro.threading_data to use a certain number of threads by @geometrikal

• New Example

  • TRPO by @jjkke88
  • Variational Autoencoder (VAE) for CelebA by @yzwxx

• New Update

  • Basic layer by @fengjiran
  • DeConv2d with auto batch size by @zsdonghao
  • tl.prepro default order=1 by @geometrikal / Ross Marchant
  • tl.files.save_npz_dict by

• Documentation by @mitar @luomai @zsdonghao

Release 1.6.1

Recommended update !

This is an incomplete list among the many features added.

• New Support

  • tl.ops.open_tb : Open TensorBoard by @Ericvulpi
  • TransposeLayer by @zsdonghao
  • InstanceNormLayer by @zsdonghao
  • tl.files.load_cyclegan_dataset : Load CycleGAN dataset by @zsdonghao

• New Examples

  • Chinese Text Anti-Spam by pakrchen
  • DAGGER - Gym Torcs , Car Racing by zsdonghao.
  • Actor-Critic for discrete action space - Cartpole, see tutorial_cartpole_ac.py

• New Updates

  • tl.ops.exit_tf by @Ericvulpi
  • Import logging in cost.py by @cclauss
  • Speed up tl.prepro.threading_data by batch processing tasks within a limited number of threads. @geometrikal

• Documentation by @zsdonghao and @shorxp

Release 1.6.0

Recommended update !

This is an incomplete list among the many features added.

• New Updates

  • New implementation of SubpixelConv2d for fast compile by @ShieLian , you can still use the old implementation via SubpixelConv2d_old
  • Fix bug of DownSampling2dLayer by @zsdonghao

• New Support

  • SubpixelConv1d for Audio Super-Resolution by @zsdonghao
  • tl.rein.choice_action_by_probs by @zsdonghao
  • tl.rein.log_weight by @zsdonghao
  • tl.files.save_ckpt and tl.files.load_ckpt by @zsdonghao

• New Examples

  • Deep Q-Network - Frozen lake, see tutorial_frozenlake_dqn.py.
  • Q-Table learning algorithm - Frozen lake, see tutorial_frozenlake_q_table.py.
  • Asynchronous Policy Gradient using TensorDB - Atari Ping Pong by nebulaV.
  • A3C for continuous action space - Bipedal Walker, see tutorial_bipedalwalker_a3c*.py.

• Good News
  TensorLayer has been accepted by ACM Multimedia 2017. If you find it is useful, please cite our paper in your project.

@article{TensorLayer2017,
   author = {Dong, Hao and Supratak, Akara and Mai, Luo and Liu, Fangde and Oehmichen, Axel and Yu, Simiao and Guo, Yike},
   journal = {ACM Multimedia},
   title = {{TensorLayer: A Versatile Library for Efficient Deep Learning Development}},
   url = {http://tensorlayer.org},
   year = {2017}
}

• Documentation by @zsdonghao