0.6.4

Made restoring Saveables a bit more efficient.
Fixed some potential memory issues.

0.6.3

This version is broken because I ran into some issues with Sonatype while trying to release it.

0.6.2

Added a check for the stack operation that avoids an ugly segfault.
Improvements to Tensor.fromByteCodable.
Minor bug fix related to the ChiefSessionCreator initialization ops.

0.6.1

Session.run no longer locks.

0.6.0

Switched to using TensorFlow 2.4.0, which required updating the handling of String tensors.
Added tf.modifyDefaultSessionConfig.
Added an API for controlling the TensorFlow logging behavior (via tf.Logging).
Fixed a bug in the gradient of tf.gather.
Lots of changes and fixes related to the implicit helpers and the way they are used when building datasets.
Refactored the handling of the native libraries and the temporary directories that they are extracted in.

0.5.7

Bug fixes:

Fixed a bug related to using tf.logSoftmax with axes set to something other than -1.
Fixed the native binaries packaging for the darwin and linux platforms.

0.5.6

Switched to using TensorFlow 2.3.1 that fixes multiple security vulnerabilities of TensorFlow 2.3.0. Also:

Fixed a minor bug related to optimizer slot variables not being saved in TF checkpoints when tf.stopGradients is used.
Switched to supporting three native platforms: linux, windows, and darwin, where the last one does not include GPU support.

0.5.5

Fix for 0.5.4 that includes pre-compiled binaries for all supported platforms.

0.5.4

Switched to using TensorFlow 2.3 an added support for Windows.

0.5.3

Accidentally broken release.

0.5.2

Accidentally broken release.

0.5.1

This release introduces support for TensorFlow 2.2 and Scala 2.13 and drops support for Scala 2.11. The distributed precompiled binaries for this version will only work with CUDA 10.1 on Linux. Finally, this release also brings improved support for implicit derivations in some cases where case classes over tensors are used.

0.5.0

This release introduces support for TensorFlow 2.0.

0.4.2

Minor release that:

Added support for a TF records reader.
Fixed a bug related to reading and writing NPY files.

0.4.1 Fixed Precompiled TF Binaries and Added Some New Features

Fixed the precompiled TensorFlow binaries, and also added the following new features:

io module:
- Added support for a new TFRecordWriter.
ops module:
- Added a new ops namespace, sparse, that includes all sparse ops.
- Added support for sparse.reorder and sparse.merge.
- Added support for parsing TF records.
- data module:
  - Added support for Dataset.shuffleAndRepeat.
- optimizers module:
  - Added support for the Adafactor optimizer.
  - Renamed SqrtDecay to RSqrtDecay which is more appropriate.
- math module:
  - Added support for batchGather.
  - Added support for bitwise ops.
- rnn module:
  - Simplified the attention mechanisms functionality so that it is now not required to tile memory tensors for beam search outside the beam search decoder.
- Moved the seq2seq module to a separate repository (that of Symphony Machine Translation).

0.4.0 More Static Data Types

This is a major release with a lot of new features related to static types for tensors and ops. The graph construction API is now statically-typed, thus enabling much better type safety than before.

Tensors and outputs are now statically-typed and the types used are the Scala types that correspond to the tensors' TensorFlow data types. For example:

val t1 = Tensor(0.5, 1) // The inferred type is Tensor[Double].
val t2 = Tensor(1, 2)   // The inferred type is Tensor[Int].
val t3 = t1 + t2        // The inferred type is Tensor[Double].
val t4 = t3.isNaN       // The inferred type is Tensor[Boolean].
val t5 = t3.any()       // Fails at compile-time because `any()` is only
                        // supported for Tensor[Boolean].

A similar situation now applies to Outputs. Ops are also typed and so is the auto-differentiation implementation.

This resulted in major simplifications in the data pipeline and the high level learn API. Datasets and dataset iterators do not "carry" T, V, D, and S types with them now, but rather just the type of the elements they contain/produce.

A new type trait called TF is also introduced that denotes supported Scala types in TensorFlow (e.g., TF[Int] and TF[Float]). Similarly, some more type traits are introduced to denote type constraints for various ops (e.g., IsIntOrUInt[Int], IsIntOrUInt[Long], IsFloatOrDouble[Float], etc.). These type traits are powered by a general implementation of union types for Scala.

Other new features include:

data module:
- Added support for the mapAndBatch transformation.

0.3.0 Static Data Types and More

With this release we have finally added support for static data type information for tensors (not for symbolic tensors yet though -- for now we effectively have support for a statically-typed version of numpy for Scala). This is an important milestone and contributes significantly to type safety, which can help catch errors at compile time, rather than runtime. For example:

val t1 = Tensor(0.5, 1) // The inferred type is Tensor[FLOAT64].
val t2 = Tensor(1, 2)   // The inferred type is Tensor[INT32].
val t3 = t1 + t2        // The inferred type is Tensor[FLOAT64].
val t4 = t3.isNaN       // The inferred type is Tensor[BOOLEAN].
val t5 = t3.any()       // Fails at compile-time because `any()` is only
                        // supported for Tensor[BOOLEAN].

Other new features include:

Improvements to the high-level learn API:
- Layers can now provide and use their own parameter generator, and can also access the current training step (using Layer.currentStep).
- Layers now support .map(...).
- Added support for batch normalization.
Added support for tf.logSigmoid and tf.lrn.
Added support for the following new metrics:
- Grouped precision.
- Precision-at-k.
data module:
- Added support for loading the extreme classification repository datasets (i.e., data.XCLoader).
- Added support for randomly splitting datasets.

0.2.4 Minor Fix

Fixed an issue with the packaged pre-compiled TensorFlow binaries that affected Linux platforms.

0.2.3 Compatibility with TensorFlow 1.9-rc1

Added compatibility with TensorFlow 1.9-rc1.

0.2.2 Pre-compiled Binaries Update

In this release we have updated the precompiled TensorFlow binaries distributed with this library.

0.2.1 Packaging Fix

In this release we have fixed an issue related to the packaging and distributing of the pre-compiled TensorFlow shared libraries.

0.2.0 Updates

In this release we have:

0.1.1 Minor Fix

This release fixes the following bugs:

Issue with the packaged pre-compiled TensorFlow binaries that affected Linux platforms.
Learn API bug where the shared name of input iterators was being set incorrectly.

I also switched to using CircleCI for continuous integration, instead of TravisCI.

0.1.0 First Official Release

This is the first official release of TensorFlow for Scala. The library website will soon be updated with information about the functionality supported by this API. Most of the main TensorFlow Python API functionality is already supported.