wekaDeeplearning4j-v1.4.0-beta.2

Changelog

• Restrict sequence iterators to sequence classifier
• Move embedding model loading to buildClassifier
• Remove zoo-models from RnnSequenceClassifier

SHA256 sums

• wekaDeeplearning4j-CPU-1.4.0-beta.2-linux-x86_64.zip
  9f765ef43c62c301106c44f8e77079761a1543c77ef7673dde4460fe1289049c
• wekaDeeplearning4j-CPU-1.4.0-beta.2-macosx-x86_64.zip
  2d49e659ca0ac3e92796ffe83d0312873da90fef03d0909c6d0331634f00aef0
• wekaDeeplearning4j-CPU-1.4.0-beta.2-windows-x86_64.zip
  2c18a457c677339b9140e814d1e29b9949142b0b7087a912d8bb4a1ef456e249
• wekaDeeplearning4j-GPU-1.4.0-beta.2-linux-x86_64.zip
  4ed190f38f80f2c09d748bb6c18d6ff61a708c6f6c1cfb61f26a7049198d6130
• wekaDeeplearning4j-GPU-1.4.0-beta.2-macosx-x86_64.zip
  18a1b363bc837d894dd791c6139f703285eb52e3cdee102edbeb082b3d04a688
• wekaDeeplearning4j-GPU-1.4.0-beta.2-windows-x86_64.zip
  b55d1e0d55af2e1ba19f005a465849998d34831e97dc694c2f955ab328056fca

wekaDeeplearning4j-v1.4.0-beta

Changelog

• Add a new classifier for sequence data: RnnSequenceClassifier (for now only applicable to text data)
• Add new Layers: LSTM, GravesLSTM, RnnOutputLayer
• Add new iterators for text data (docs):
  • TextEmbeddingInstanceIterator
  • TextFilesEmbeddingInstanceIterator
• Make epoch evaluations in EpochListener optional
• Add dataset CacheMode for performance improvements (docs)
• Various fixes

wekaDeeplearning4j-v1.3.4

Changes

• Update package naming scheme to wekaDeeplearning4j-<BACKEND>-<PLATFORM>
• Make use of OSName and OSArch property in the weka package management
• Add all conflicting packages to the Precludes property on build time
• Fix log line separator on windows

Attention
Users with a Weka version of 3.8.0 need to manually uninstall the old wekaDeeplearning4j packages since 3.8.0 does not support the Precludes property which automatically checks for incompatible packages.

wekaDeeplearning4j-v1.3.3

Changes

• Critical bug fix: The number of updates was dependent on the setup of EpochListener due to a missed reset() call on the training iterator at each epoch

wekaDeeplearning4j-v1.3.2

Changes

• Fix exception when setting step size (N) for evaluations in EpochListener to zero
• (Critical) Fix memory consumption by switching from dl4j WorkspaceMode.SINGLE to WorkspaceMode.SEPARATE

wekaDeeplearning4j-v1.3.1

Changes

• Choosing a zoo model now adds its layers to the current configuration. It is also possible to change the layer setup and parameters (keep in mind, that this will reset the zooModel option to CustomNet since the changed setup does not comply with the actually chosen zoo model anymore)
• Add warning if validation set in EarlyStopping was too small
• Add global information to different options in the GUI
• Add more warnings if invalid setups which configuring the network
• Fix log file if WEKA_HOME was not set
• Add layer configuration validation
• Add option for evaluation after N epochs to EpochListener
• Reduce evaluation time

Update

This version suffered from a memory bug which is fixed in v1.3.2 and above.

wekaDeeplearning4j-v1.3.0

Changes

• Update documentation
• Use IUpdater backend
• Remove per-layer parameters
• Introduce EarlyStopping condition (see the documentation)
• Use $WEKA_HOME/network.log as default logging file
• Fix loading images for directories with spaces
• Logs now show metrics for training and validation set by default if EarlyStopping is enabled

Attention

This version is not compatible with configurations of earlier versions (< 1.3.0) due to the removal of many parameters in the layer configuration. These parameters are now in the NeuralNetConfiguration. Furthermore, the Updater parameters have been moved to the updater-editor. These changes make the use of many parameters more clear.

Update

Downloads removed due to a bug. Please use v1.3.1 or later.

wekaDeeplearning4j-v1.2.3

Changes

• Fix memory allocation in zoo-models by using Dl4j's workspaces
• Use ComputationGraphs instead of MultiLayerNetworks

wekaDeeplearning4j-v1.2.2

Changes

• Fix windows paths
• Enable batches at inference time
• Enable modelzoo

wekaDeeplearning4j-v1.2.1

Changes

• Bug fixes