# 1.0.0-beta5

## Highlights - 1.0.0-beta5 Release

* Added model server - remote inference of SameDiff and DL4J models using JSON or (optionally) binary serialization
  * Server: See [JsonModelServer](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/main/java/org/deeplearning4j/remote/JsonModelServer.java)
  * Client: See [JsonRemoteInference](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-remote/nd4j-json-client/src/main/java/org/nd4j/remote/clients/JsonRemoteInference.java)
  * Tests/examples: See [Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/JsonModelServerTest.java) and [Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/BinaryModelServerTest.java)
* Added Scala 2.12 support, dropped Scala 2.10 support. Modules with Scala dependencies are now released with Scala 2.11 and 2.12 versions
* Apache Spark 1.x support dropped (now only Spark 2.x is supported). Note: Spark version suffix dropped: For upgrading: `1.0.0-beta4_spark2 -> 1.0.0-beta5`
* Added FastText support to deeplearning4j-nlp
* CUDA support for all ND4J/SameDiff Operations
  * In 1.0.0-beta4, some operations were CPU only. Now, all operations have full CUDA support
* Added support for new data types in ND4J (and DL4J/SameDiff): BFLOAT16, UINT16, UINT32, UINT64
* ND4J: Implicit broadcasting support added to INDArray (already present in SameDiff - for example shape `[3,1]+[3,2]=[3,2]`)
* CUDA 9.2, 10.0 and 10.1-Update2 still supported
  * NOTE: For CUDA 10.1, CUDA 10.1 update 2 is recommended. CUDA 10.1 and 10.1 Update 1 will still run, but rare internal cuBLAS issues may be encountered in heavily multi-threaded code on some systems
* Dependency upgrades: Jackson (2.5.1 to 2.9.9/2.9.9.3), Commons Compress (1.16.1 to 1.18), Play Framework (2.4.8 to 2.7.3), Guava: (20.0 to 28.0-jre, and shaded to avoid dependency clashes)
* CUDA: now host (RAM) buffers are only allocated when required (previously: host buffers were always allocated), in addition to device (GPU) buffer

## Deeplearning4J

### Deeplearning4J: Features and Enhancements

* Added FastText - inference and training, including OOV (out of vocabulary) support ([Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-nlp-parent/deeplearning4j-nlp/src/main/java/org/deeplearning4j/models/fasttext/FastText.java))
* Scala 2.12 support added, Scala 2.10 support dropped ([Link](https://github.com/SkymindIO/deeplearning4j/pull/199))
* Added model server (DL4J and SameDiff models, JSON and binary communication) - [JsonModelServer](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/main/java/org/deeplearning4j/remote/JsonModelServer.java), [JsonRemoteInference](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-remote/nd4j-json-client/src/main/java/org/nd4j/remote/clients/JsonRemoteInference.java), [Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/JsonModelServerTest.java), [Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/BinaryModelServerTest.java)
* Added saved model format validation utilities - DL4JModelValidator, DL4JKerasModelValidator ([Link](https://github.com/eclipse/deeplearning4j/pull/7701))
* Added LabelLastTimeStepPreProcessor ([Link](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/dataset/api/preprocessor/LabelLastTimeStepPreProcessor.java))
* BertIterator: added option to prepend token to the output (such as `[cls]` expected by some models) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/39))
* Added trace level logging to MultiLayerNetwork and ComputationGraph assist with debugging certain issues ([Link](https://github.com/SkymindIO/deeplearning4j/pull/79))
* Upsampling3D: Added NDHWC support ([Link](https://github.com/eclipse/deeplearning4j/issues/8016))
* MergeVertex now supports broadcasting ([Link](https://github.com/eclipse/deeplearning4j/issues/6488))
* LSTM and Dropout will now fall back on built-in implementations if an exception is encountered from cuDNN (same as Subsampling/ConvolutionLayer) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/152))
* Improved JavaDoc and cleanup up API for WordVectorSerializer ([Link](https://github.com/SkymindIO/deeplearning4j/pull/221), [Link](https://github.com/eclipse/deeplearning4j/issues/8137))

### Deeplearning4J: Bug Fixes and Optimizations

* Updated deeplearning4j-ui theme ([Link](https://github.com/eclipse/deeplearning4j/pull/7683))
* Fixed an issue with MergeVertex and CNN3D activations ([Link](https://github.com/eclipse/deeplearning4j/issues/7715))
* Fixed typo in Yolo2OutputLayer builder/configuration method name ([Link](https://github.com/eclipse/deeplearning4j/pull/7728))
* Improved ComputationGraph builder InputType validation ([Link](https://github.com/eclipse/deeplearning4j/pull/7752))
* Removed dl4j-spark-ml module until it can be properly maintained ([Link](https://github.com/eclipse/deeplearning4j/pull/7764))
* Fixed an issue with BertWordPieceTokenizerFactory and bad character encoding ([Link](https://github.com/eclipse/deeplearning4j/issues/7678))
* Fixed an issue with LearnedSelfAttentionLayer and variable minibatch size ([Link](https://github.com/eclipse/deeplearning4j/pull/7780), [Link](https://github.com/eclipse/deeplearning4j/issues/7777))
* Fixed issue with SharedTrainingMaster controller address when set from environment variable ([Link](https://github.com/eclipse/deeplearning4j/issues/7786))
* Fixed issue with SameDiffOutputLayer initialization under some circumstances ([Link](https://github.com/eclipse/deeplearning4j/issues/7785))
* https is now used by default for data and zoo model downloads ([Link](https://github.com/eclipse/deeplearning4j/issues/7779), [Link](https://github.com/eclipse/deeplearning4j/pull/7793))
* Fixed an issue where UI WebJars dependencies would check for updates on every single build ([Link](https://github.com/eclipse/deeplearning4j/issues/7730), [Link](https://github.com/eclipse/deeplearning4j/pull/7793))
* Fixed issue where Upsampling layer memory report could produce an OOM exception ([Link](https://github.com/eclipse/deeplearning4j/pull/8015))
* Improved UX/validation for RecordReaderDataSetIterator ([Link](https://github.com/eclipse/deeplearning4j/pull/8042))
* Fixed an issue where EmbeddingSequenceLayer would not check mask array datatype ([Link](https://github.com/eclipse/deeplearning4j/issues/7866))
* Improved validation when initializing networks with a non rank-2 (shape \[1, numParams]) array ([Link](https://github.com/eclipse/deeplearning4j/issues/7890))
* Fixed a DataType issue for BertIterator ([Link](https://github.com/SkymindIO/deeplearning4j/pull/17))
* Fixed Word2Vec model backward compatibilty (beta3 and earlier models now loadable again) [Link](https://github.com/SkymindIO/deeplearning4j/pull/18)
* Fixed issue where some Keras import models could fail with `Could not read abnormally long HDF5 attribute` ([Link](https://github.com/eclipse/deeplearning4j/issues/7919))
* Added validation for RnnOutputLayer - feature/label array lengths ([Link](https://github.com/eclipse/deeplearning4j/issues/7925))
* Fixed an issue where SameDiffOutputLayer would not support variable minibatch size ([Link](https://github.com/eclipse/deeplearning4j/issues/7844))
* Fixed DL4J SameDiff layer mask support ([Link](https://github.com/SkymindIO/deeplearning4j/pull/67))
* DL4J UI: Fixed an issue where tab switching did not work when visualizing saved/stored data ([Link](https://github.com/eclipse/deeplearning4j/issues/7954), [Link](https://github.com/SkymindIO/deeplearning4j/pull/243))
* DL4J UI: Fixed a rare UI threading issue ([Link](https://github.com/eclipse/deeplearning4j/issues/8017))
* Fixed a Keras import issue with JSON format change ([Link](https://github.com/eclipse/deeplearning4j/issues/7992))
* Fixed a Keras import issue where updater learning rate schedule could be imported incorrectly ([Link](https://github.com/SkymindIO/deeplearning4j/pull/84))
* Fixed an issue with CnnSentenceDataSetIterator when using `UnknownWordHandling.UseUnknownVector` ([Link](https://github.com/eclipse/deeplearning4j/issues/8121), [Link](https://github.com/eclipse/deeplearning4j/issues/8120))
* Fixes and optimizations to DL4J SameDiff layers ([Link](https://github.com/SkymindIO/deeplearning4j/pull/156))
* MultiLayerNetwork/ComputationGraph will now log the original exception if a second exception occurs during workspace closing, instead of swallowing it (inference/fit operation try/finally blocks) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/161))
* Upgraded dependencies: Jackson (2.5.1 to 2.9.9/2.9.9.3), Commons Compress (1.16.1 to 1.18), Play Framework (2.4.8 to 2.7.3), Guava: (20.0 to 28.0-jre, shaded to avoid dependency clashes) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/199))
* Logging framework can now be configured for DL4J UI (due to Play framework dependency upgrade) ([Link](https://github.com/eclipse/deeplearning4j/issues/3808))
* Reduced amount of garbage produced by MnistDataFetcher (impacts MNIST and EMNIST DataSetIterators) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/200))
* Activation function backpropagation has been optimized for many activation functions ([Link](https://github.com/SkymindIO/deeplearning4j/pull/211), [Link](https://github.com/SkymindIO/deeplearning4j/pull/207))

### Deeplearning4j: Transition Guide, 1.0.0-beta4 to 1.0.0-beta5

* DL4J AsyncDataSetIterator and AsyncMultiDataSetIterator moved to ND4J, use `org.nd4j.linalg.dataset.Async(Multi)DataSetIterator` instead
* Saved models with custom layers from 1.0.0-alpha and before can no longer be loaded. Workaround: load in 1.0.0-beta4, and re-save the model ([Link](https://github.com/SkymindIO/deeplearning4j/pull/199)). Models without custom layers can still be loaded back to 0.5.0
* Apache Spark 1.x support dropped (now only Spark 2.x is supported). Note: Spark version suffix dropped: For upgrading, change versions as follows: `1.0.0-beta4_spark2 -> 1.0.0-beta5`
* Scala 2.10 dropped, Scala 2.12 added (for modules with Scala dependencies)

### Deeplearning4j: 1.0.0-beta5 Known Issues

* dl4j-spark\_2.11 and \_2.12 dependencies incorrectly pull in datavec-spark\_2.11/2.12 version `1.0.0-SNAPSHOT`. Workaround: control version using dependency management as per [here](https://github.com/eclipse/deeplearning4j-examples/pull/901/files) or [here](https://gist.github.com/AlexDBlack/5721cb6fbd6cd98b6702cd49e733dacf)
* Some layers (such as LSTM) may run slower on 1.0.0-beta5 than 1.0.0-beta4 on CUDA when not using cuDNN, due to added synchronization. This synchronization will be removed in the next release after 1.0.0-beta5
* CUDA 10.1: Rare internal cuBLAS issues may be encountered in heavily multi-threaded code on some systems, when running CUDA 10.1 Update 1 (and maybe 10.1). CUDA 10.1 update 2 is recommended.

## ND4J and SameDiff

### ND4J/SameDiff: Features and Enhancements

* Added new data types: BFLOAT16, UINT16, UINT32, UINT64 ([Link](https://github.com/eclipse/deeplearning4j/pull/7723))
* CUDA support for all operations without CUDA implementations ([Link](https://github.com/SkymindIO/deeplearning4j/pull/26), [Link](https://github.com/SkymindIO/deeplearning4j/pull/57), [Link](https://github.com/SkymindIO/deeplearning4j/pull/63), [Link](https://github.com/SkymindIO/deeplearning4j/pull/69), [Link](https://github.com/SkymindIO/deeplearning4j/pull/95))
* Added model server (DL4J and SameDiff models, JSON and binary communication) - [JsonModelServer](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/main/java/org/deeplearning4j/remote/JsonModelServer.java), [JsonRemoteInference](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-remote/nd4j-json-client/src/main/java/org/nd4j/remote/clients/JsonRemoteInference.java), [Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/JsonModelServerTest.java), [Link](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/BinaryModelServerTest.java)
* Added support for empty arrays with zeros in shape, for compatibility with TensorFlow import ([Link](https://github.com/SkymindIO/deeplearning4j/pull/10))
* CUDA: now host (RAM) buffers are only allocated when required (previously: host buffers were always allocated), in addition to device (GPU) buffer
* Improved SameDiff training API - added "in line" test set evaluation, returning History object with loss curve, etc ([Link](https://github.com/SkymindIO/deeplearning4j/pull/99))
* Added saved model format validation utilities - Nd4jValidator, Nd4jCommonValidator ([Link](https://github.com/eclipse/deeplearning4j/pull/7701))
* Added SameDiff ScoreListener (equivalent to DL4J ScoreIterationListener/PerformanceListener) ([Link](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/autodiff/listeners/impl/ScoreListener.java), [Link](https://github.com/eclipse/deeplearning4j/pull/7774))
* Added SameDiff.convertDataTypes method, for variable dtype conversion ([Link](https://github.com/eclipse/deeplearning4j/blob/a76a44e1988fa785f38fd99df17e7396a7746b53/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/autodiff/samediff/SameDiff.java#L3799-L3812))
* Added crop and resize op ([Link](https://github.com/eclipse/deeplearning4j/pull/7741))
* DL4J AsyncDataSetIterator and AsyncMultiDataSetIterator moved to ND4J [Link](https://github.com/eclipse/deeplearning4j/pull/7774)
* Added basic/MVP SameDiff UI listener ([Link](https://github.com/eclipse/deeplearning4j/pull/7787))
* Added SameDiff CheckpointListener ([Link](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/autodiff/listeners/checkpoint/CheckpointListener.java), [Link](https://github.com/eclipse/deeplearning4j/pull/7807))
* Added SameDiff name scopes ([Link](https://github.com/eclipse/deeplearning4j/blob/a76a44e1988fa785f38fd99df17e7396a7746b53/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/autodiff/samediff/SameDiff.java#L544-L579))
* SameDiff: Updater state and training configuration is now written to FlatBuffers format ([Link](https://github.com/eclipse/deeplearning4j/pull/7825))
* Added c++ benchmark suite callable from Java - call using `Nd4j.getExecutioner().runLightBenchmarkSuit()` and `Nd4j.getExecutioner().runFullBenchmarkSuit()` ([Link](https://github.com/SkymindIO/deeplearning4j/pull/3))
* Added SameDiff.save/load methods with InputStream/OutputStream arguments ([Link](https://github.com/eclipse/deeplearning4j/issues/7856), [Link](https://github.com/SkymindIO/deeplearning4j/pull/6))
* Added axis configuraiton for evaluation instances (Evaluation, RegressionEvaluation, ROC, etc - getAxis and setAxis methods) to allow different data formats (NCHW vs. NHWC for CNNs, for example) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/6))
* SameDiff: Added support to convert constants to placeholders, via SDVariable.convertToConstant() method ([Link](https://github.com/SkymindIO/deeplearning4j/pull/11))
* SameDiff: Added GradCheckUtil.checkActivationGradients method to check activation gradients for SameDiff instance (not just parameter gradients as in existing gradient check methods) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/19))
* Added CheckNumerics op ([Link](https://github.com/SkymindIO/deeplearning4j/pull/23/commits/8399ef0e11d7a23a3636516fab7ee6fdc7d8f313))
* Added FakeQuantWithMinMaxArgs and FakeQuantWithMinMaxVars ops ([Link](https://github.com/SkymindIO/deeplearning4j/pull/24))
* Added INDArray reduction methods with "keep dimensions" option - for example, `INDArray.mean(boloean, int... dimension)` ([Link](https://github.com/SkymindIO/deeplearning4j/pull/33))
* Added Nd4j SystemInfo class - SystemInfo.getSystemInfo, .writeSystemInfo(File) to aid with debugging issues ([Link](https://github.com/SkymindIO/deeplearning4j/pull/34), [Link](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/systeminfo/SystemInfo.java))
* Added INDArray.toString(NDArrayStrings options), toStringFull() and toString overloads for easier control of array printing ([Link](https://github.com/SkymindIO/deeplearning4j/pull/36))
* Added HashCode op, INDArray.hashCode() ([Link](https://github.com/SkymindIO/deeplearning4j/pull/50))
* SameDiff: added whileLoop, ifCond methods for loops/conditional ops ([Link](https://github.com/SkymindIO/deeplearning4j/pull/52))
* Cleaned up some infrequently used Nd4j methods ([Link](https://github.com/SkymindIO/deeplearning4j/pull/89), [Link](https://github.com/SkymindIO/deeplearning4j/pull/101), [Link](https://github.com/SkymindIO/deeplearning4j/pull/102), [Link](https://github.com/SkymindIO/deeplearning4j/pull/231))
* Added bitwise integer operations: left/right bit shift, left/right cyclical bit shift, bitwise Hamming distance ([Link](https://github.com/SkymindIO/deeplearning4j/pull/115), [Link](https://github.com/SkymindIO/deeplearning4j/pull/118), [Link](https://github.com/SkymindIO/deeplearning4j/pull/126), [Link](https://github.com/SkymindIO/deeplearning4j/pull/192), [Link](https://github.com/SkymindIO/deeplearning4j/pull/195))
* deeplearning4j-nlp: renamed AggregatingSentencePreProcessor to sentencePreProcessor method ([Link](https://github.com/eclipse/deeplearning4j/issues/8122))
* Upgraded (and shaded) Protobuf version - 3.5.1 to 3.8.0 ([Link](https://github.com/SkymindIO/deeplearning4j/pull/162))
* Switched to c=style error handling for libnd4j native operations ([Link](https://github.com/SkymindIO/deeplearning4j/pull/169))
* Renamed FlatBuffers enum `org.nd4j.graph.DataType` to `org.nd4j.graph.DType` to avoid users importing incorrect type when using Nd4j methods ([Link](https://github.com/SkymindIO/deeplearning4j/pull/228), [Link](https://github.com/eclipse/deeplearning4j/issues/8159))
* Added SameDiff.bitwise namespace for bitwise ops ([Link](https://github.com/SkymindIO/deeplearning4j/pull/232), [Link](https://github.com/eclipse/deeplearning4j/blob/master/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/autodiff/samediff/ops/SDBitwise.java))

### ND4J/SameDiff: Bug Fixes and Optimizations

* Updated to JavaCPP/JavaCV 1.5.1-1 ([Link](https://github.com/eclipse/deeplearning4j/pull/8004))
* SameDiff: Placeholders must now only be provided if required to calculate the requested variables ([Link](https://github.com/eclipse/deeplearning4j/pull/7814))
* SameDiff: Fixed an issue with duplicate variable name validation ([Link](https://github.com/eclipse/deeplearning4j/issues/7705))
* SameDiff: Fixed an issue with SDVariable.getArr for scalars ([Link](https://github.com/eclipse/deeplearning4j/issues/7546))
* Added delayed mode to DeviceLocalNDArray (don't replicate to device until needed) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/149))
* ND4J: Fixed an issue with writing 0d (scalar) NDArrays in numpy .npy format ([Link](https://github.com/eclipse/deeplearning4j/pull/7712))
* Fixed an issue with Pad operation for some constant cases ([Link](https://github.com/eclipse/deeplearning4j/pull/7713))
* Fixed some issues with strided\_slice operation ([Link](https://github.com/eclipse/deeplearning4j/pull/7719), [Link](https://github.com/eclipse/deeplearning4j/pull/7823), [Link](https://github.com/SkymindIO/deeplearning4j/pull/14))
* SameDiff: Fixed issue with DataType inference for some ops using ND4J default datatype ([Link](https://github.com/eclipse/deeplearning4j/pull/7699))
* INDArray.castTo(DataType) is now a no-op when array is already the correct type ([Link](https://github.com/eclipse/deeplearning4j/issues/7754))
* SameDiff: Fixed an issue with training mixed precision networks ([Link](https://github.com/eclipse/deeplearning4j/issues/6992))
* Fixed an issue where Evaluation class was incorrectly reporting macro-averaged precision for binary case ([Link](https://github.com/eclipse/deeplearning4j/issues/7802))
* Removed trainableParams config/field from SameDiff TrainingConfig (no longer required) ([Link](https://github.com/eclipse/deeplearning4j/pull/7807))
* Improvements and cleanup to ND4J Javadoc ([Link](https://github.com/eclipse/deeplearning4j/pull/7997), [Link](https://github.com/SkymindIO/deeplearning4j/pull/110), [Link](https://github.com/SkymindIO/deeplearning4j/pull/111), [Link](https://github.com/SkymindIO/deeplearning4j/pull/112))
* Fixed an issue with Cholesky Lapack op on CUDA ([Link](https://github.com/eclipse/deeplearning4j/pull/8188), [Link](https://github.com/eclipse/deeplearning4j/issues/8167))
* Fixed an issue where \[1,N] and \[N,1] arrays were not considered a matrix (rank 2 array) according to INDArray.isMatrix() ([Link](https://github.com/eclipse/deeplearning4j/issues/7839))
* Fixed RegressionEvaluation for 4D arrays (CNNs / segmentation) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/6), [Link](https://github.com/eclipse/deeplearning4j/issues/7859))
* Fixed issue with INDArray.median(int... dimension) ([Link](https://github.com/eclipse/deeplearning4j/issues/7848))
* Fixed NPE that could occur when executing gather operation backprop ([Link](https://github.com/eclipse/deeplearning4j/issues/7852))
* Fixed issue with LogSumExp operation Java/C++ mapping ([Link](https://github.com/eclipse/deeplearning4j/issues/7850))
* Added header validation when reading Numpy .npy files, to ensure file is valid ([Link](https://github.com/SkymindIO/deeplearning4j/pull/46))
* Fixed a possible issue with reading Numpy .npy files on CUDA ([Link](https://github.com/SkymindIO/deeplearning4j/pull/64))
* Fixed an issue when reading Numpy .npy boolean files ([Link](https://github.com/SkymindIO/deeplearning4j/pull/91))
* Various fixes for TensorFlow import ([Link](https://github.com/SkymindIO/deeplearning4j/pull/55))
* Fixed an issue with a small number of Nd4j.create methods not creating arrays corresponding to the java primitive ([Link](https://github.com/eclipse/deeplearning4j/issues/8057))
* Improved shape validation for some Nd4j.create methods ([Link](https://github.com/eclipse/deeplearning4j/issues/7551))
* Cleaned up unmaintained Nd4j.createSparse methods ([Link](https://github.com/SkymindIO/deeplearning4j/pull/92))
* Fixed a CUDA issue for CUDA GPUs with CC 3.0 ([Link](https://github.com/SkymindIO/deeplearning4j/pull/105))
* Fixed some possible integer overflows in c++ code ([Link](https://github.com/SkymindIO/deeplearning4j/pull/108))
* Removed deprecated methods: Nd4j.trueScalar and Nd4j.trueVector ([Link](https://github.com/SkymindIO/deeplearning4j/pull/129), [Link](https://github.com/SkymindIO/deeplearning4j/pull/145))
* Fixed an issue where some JVMs could warn about "Illegal reflective access" due to a (now removed) SameDiff dependency ([Link](https://github.com/eclipse/deeplearning4j/issues/8123))
* SDVariable now no longer extends DifferentialFunction ([Link](https://github.com/SkymindIO/deeplearning4j/pull/150))
* Moved numerous operation calculateOutputShape instances from Java to C++ ([Link](https://github.com/SkymindIO/deeplearning4j/pull/151))
* Fixed an issue where maxpool2d\_bp could throw an exception when NaN values are present ([Link](https://github.com/SkymindIO/deeplearning4j/pull/160))
* Fixed an issue with concatenation of empty shapes (with zeros) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/167))
* Removed INDArray.javaTensorAlongDimension ([Link](https://github.com/SkymindIO/deeplearning4j/pull/170))
* LayerNorm operation now properly supports axis arg, NCHW format data ([Link](https://github.com/eclipse/deeplearning4j/issues/8008))
* libnd4j: cuBLAS hgemm (FP16 gemm) wil only be called for devices with compute capability >= 5.3 due to cuBLAS limitations ([Link](https://github.com/SkymindIO/deeplearning4j/pull/181))
* Nd4j.readNumpy optimized ([Link](https://github.com/SkymindIO/deeplearning4j/pull/183))
* Added configurable alpha parameter to ELU and lrelu\_bp operations in c++ ([Link](https://github.com/SkymindIO/deeplearning4j/pull/213))
* Cleaned up SameDiff SDCNN/SDRNN (SameDiff.cnn, .rnn) API/methods ([Link](https://github.com/SkymindIO/deeplearning4j/pull/230), [Link](https://github.com/SkymindIO/deeplearning4j/pull/238))

### ND4J: Transition Guide, 1.0.0-beta4 to 1.0.0-beta5

* OldAddOp, OldSubOp, etc removed: Replace with AddOp, SubOp, etc
* Nd4j.trueScalar and trueVector removed; use Nd4j.scalar and Nd4j.createFromArray methods
* INDArray.javaTensorAlongDimension removed; use INDArray.tensorAlongDimension instead
* INDArray.lengthLong() removed; use INDArray.length() instead

### ND4J: 1.0.0-beta5 Known Issues

* nd4j-native on some OSX systems can fail with `Symbol not found: ___emutls_get_address` - See [this link](https://github.com/eclipse/deeplearning4j/issues/8217)
* SBT 1.3.0 can fail with an `Illegal character in path` error; SBT 1.2.8 is OK. This is an SBT issue, not an ND4J issue. See [this link](https://github.com/sbt/sbt/issues/5046) for details

## DataVec

### DataVec: Features and Enhancements

* ImageRecordReader: Support for 16-bit TIFF added ([Link](https://github.com/eclipse/deeplearning4j/pull/7731))
* Added SequenceTrimToLengthTransform ([Link](https://github.com/SkymindIO/deeplearning4j/pull/61))

### DataVec: Bug Fixes and Optimizations

* Fixed an issue with AnalyzeSpark and String columns ([Link](https://github.com/eclipse/deeplearning4j/issues/7680))
* Fixed an issue with URL scheme detection in NumberedFileInputScheme ([Link](https://github.com/eclipse/deeplearning4j/pull/7742))
* Fixed an issue with RandomPathFilter sampling being biased ([Link](https://github.com/eclipse/deeplearning4j/pull/7769), [Link](https://github.com/eclipse/deeplearning4j/issues/7738))

## RL4J

### RL4J: Features and Enhancements

* API cleanup and refactoring ([Link](https://github.com/eclipse/deeplearning4j/pull/7958), [Link](https://github.com/eclipse/deeplearning4j/pull/8034), [Link](https://github.com/eclipse/deeplearning4j/pull/8050), [Link](https://github.com/eclipse/deeplearning4j/pull/8065))

### RL4J: Bug Fixes and Optimizations

* Fixed issue with compression for HistoryProcessor ([Link](https://github.com/eclipse/deeplearning4j/pull/7819))

## Arbiter

### Bug Fixes and Optimizations

* Updated EvaluationScoreFunction to use ND4J Evaluation class metrics ([Link](https://github.com/eclipse/deeplearning4j/issues/7804))
* Fixed incorrect search size in GridSearchCandidateGenerator ([Link](https://github.com/eclipse/deeplearning4j/issues/8082))

### Arbiter: Known Issues

* The Jackson version upgrade necessitated a change to how generic object serialization was performed; Arbiter JSON data stored in 1.0.0-beta4 or earlier format may not be readable in 1.0.0-beta5 ([Link](https://github.com/SkymindIO/deeplearning4j/pull/237))

## ND4S

### ND4S Features and Enhancements

* Added full data type support to ND4S as per ND4J ([Link](https://github.com/SkymindIO/deeplearning4j/pull/51))
* Added syntactic sugar for SameDiff (implicits, operator overloads) ([Link](https://github.com/SkymindIO/deeplearning4j/pull/113))