TensorFlow Addons v0.5.1

Release Notes

• Built against: tensorflow-gpu==2.0.0-rc1 (Linux); tensorflow==2.0.0-rc1 (Mac)
• Automatically install the correct tensorflow version during the installation.
• Update document for CorrelationCost class, and hide correlation_cost function.

Known problem

• #532 Import crashed on a non-GPU Linux system. Workaround: please uninstall tensorflow-gpu, and then install tensorflow-cpu manually instead.

TensorFlow Addons v0.5.0

Release Notes

• Built against: tensorflow-gpu==2.0.0-rc0 (Linux); tensorflow==2.0.0-rc0 (Mac)
• GPU Kernels now part of PyPi package (Linux)
• API docs found here in repository

Changelog

tfa.activations

• Add GeLu (#424, #427)

tfa.image

• Add sparse image warp (#252)
• Support batch input for median_filter2d (#288)
• Speed up median_filter2d (#304)
• Add Translate ops (#320)
• Fix unknown ranks bug, image is compatible with tf.data now (#330)
• Add Connected components (#409)

tfa.layers

• Add correlation cost layer (#207)
• WeightNormalization is compatible with tf.functions now (#458)

tfa.losses

• Fix compile triplet loss within keras model (#298)
• Add npairs_loss (#309)

tfa.metrics

• Add R-square metric (#283)
• Add F1 metric - micro/macro/weighted (#284)
• F1-Beta Score (#315)
• Multilabel Confusion Matrix (#324)

tfa.seq2seq

• Improve API for resetting AttentionMechanism memory (#354)
• Minor bugfixs and improvement

tfa.text

• Add conditional random fields (#314)

Examples

• Normalizations (#144)
• TripletLoss (#360)
• Image Ops (#398, #444)

Thanks to our Contributors

Aakash Kumar Nain, Dheeraj R Reddy, Fei Hu, Guillaume Klein, Karmel Allison, Keshan, Kyle Beauchamp, Matt Elsey, Moritz Kröger, parth-p, Patrick Wieschollek, Qianli Scott Zhu, Saishruthi, Sayooj, Sean Morgan, seanccho, Sebastian J. Mielke, Sergii Khomenko, Tzu-Wei Sung, Yan Facai (颜发才)

TensorFlow Addons v0.4.0

Release Notes

• Built against: tensorflow==2.0.0-beta1
• API docs found in the repository

Changelog

• Assorted bugfixes throughout
• Cleanup to use public tensorflow APIs throughout

tfa.image

• Add mean filter 2d (#185)
• Speed up mean_filter2d with depthwise_conv2d (#235)

tfa.metrics

• Add Cohen's Kappa (#267)

tfa.optimizers

• Bugfix correctly subclass Adam in LazyAdam optimizer (#261)
• Set epsilon as a constant in LazyAdam optimizer (#248)

tfa.rnn

• Pass kwargs to wrapped cell in AttentionWrapper (#272)

Thanks to our Contributors

• Aakash Kumar Nain
• Dheeraj R Reddy
• Guillaume Klein
• Kyle Beauchamp
• Mainak Dutta
• Qianli Scott Zhu
• Sean Morgan
• Shashvat Chand Shahi
• Tzu-Wei Sung
• Yan Facai (颜发才)

TensorFlow Addons v0.3.1

Release Notes

• Built against: tensorflow==2.0.0.a0
• MacOS & Linux whl files are now available on pypi
• New subpackage tfa.RNN
• API docs have been generated and will soon be available on the website
• Addons is now up to 9 subpackage and submodule maintainers!

Changelog

• Assorted bugfixes throughout

tfa.image

• Add median filter 2d
• Add dense image warp
• Add euclidean distance transform
• Add rotate transform

tfa.layers

• Remove LayerNormalization as it's now part of TF Core

tfa.losses

• Add contrastive loss

tfa.optimizers

• Add decoupled weight decay optimizers (SGDW, AdamW)
• Add moving average optimizer

tfa.rnn

• Add NASCell
• Add LayerNormLSTMCell

tfa.seq2seq

• API and other fixes

Examples

• Add example for lazy adam optimizer
• Add example for weight normalization layer

Thanks to our Contributors

• Armando Fandango
• Aurélien Geron
• Dheeraj R Reddy
• Gokkul Nath
• Kyle Beauchum
• Mainak Dutta
• Moritz Kröger
• PhilJd
• Qianli Scott Zhu
• Sean Morgan
• Shreyas Sharma
• Tzu-Wei Sung
• Yan Facai (颜发才)

TensorFlow Addons 0.2.1

Release Notes

• Rebuild package with gcc4.8 as per issues seen here
• Remove python3.7 package until tool chain is finalized

TensorFlow Addons 0.2.0

Release Notes

• Build against: tensorflow==2.0.0-alpha0
• Restructure into subpackages for future maintainership
• Add python3.7 package

Change Log

Seq2seq

• Add new subpackage seq2seq to TensorFlow Addons

Layers

• Add LayerNormalization
• Add GroupNormalization
• Add InstanceNormalization

Losses

• Add focal_loss

Optimizers

• Rename LazyAdamOptimizer to LazyAdam in order to match TF2 standards

Image

• Add adjust_hsv_in_yiq (ported from tf.contrib)
• Add random_hsv_in_yiq (ported from tf.contrib)

Thank You to All Our Contributors

• Aakash Kumar Nain
• Amit Patankar
• Dheeraj R Reddy
• Jakub Arnold
• Jeremiah Harmsen
• Kapil Sachdeva
• Karmel Allison
• Moritz Kröger
• Paige Bailey
• Qianli Scott Zhu
• Tzu-Wei Sung
• Yan Facai (颜发才)

TensorFlow Addons 0.1.1

Release Notes:

• Reverted image op names to use V2 suffix in order to preserve compatibility with previous saved models.
• Workaround fix for test utils until tensorflow/tensorflow#26723 is fixed

TensorFlow Addons 0.1.0

Release Notes

This is the first release of TensorFlow Addons!

Change Log

Activations

• Add Sparsemax (ported from tf.contrib)

Image

• Add transform (ported from tf.contrib)

Layers

• Add Maxout (ported from tf.contrib)
• Add PoincareNormalize (ported from tf.contrib)
• Add WeightNormalization

Losses

• Add LiftedStructLoss (ported from tf.contrib)
• Add SparsemaxLoss (ported from tf.contrib)
• Add TripletSemiHardLoss (ported from tf.contrib)

Optimizers

• Add LazyAdamOptimizer (ported from tf.contrib)

Text

• Add skip_gram_sample (ported from tf.contrib)

Thank You to All Our Contributors

Those included in the release

• Yan Facai (颜发才)
• Tabish Shaikh
• Andreas Madsen
• Lukas Geiger

Those waiting for PRs to be merged

• Moritz Kröger
• Aakash Kumar Nain
• Qianli Scott Zhu
• Tzu-Wei Sung

We also want to extend a thank you to the Google team members who have helped with CI setup and reviews!