layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE || !input.isAttached() ? input : input.migrate(); List<INDArray> activations = new ArrayList<>(); MemoryWorkspace workspace = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SINGLE ? Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceExternal) : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( if (layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SEPARATE) Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceFeedForward).initializeWorkspace();
/** Calculate the output of the network, with masking arrays. The masking arrays are used in situations such * as one-to-many and many-to-one recurrent neural network (RNN) designs, as well as for supporting time series * of varying lengths within the same minibatch. */ public INDArray output(INDArray input, boolean train, INDArray featuresMask, INDArray labelsMask) { WorkspaceMode cMode = layerWiseConfigurations.getTrainingWorkspaceMode(); layerWiseConfigurations.setTrainingWorkspaceMode(layerWiseConfigurations.getInferenceWorkspaceMode()); MemoryWorkspace workspace = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal); try (MemoryWorkspace wsE = workspace.notifyScopeEntered()) { INDArray ret = silentOutput(input, train, featuresMask, labelsMask).detach(); layerWiseConfigurations.setTrainingWorkspaceMode(cMode); return ret; } }
/** * Label the probabilities of the input * * @param input the input to label * @param train whether the output * is test or train. This mainly * affect hyper parameters such as * drop out where certain things should * be applied with activations * @return a vector of probabilities * given each label. * <p> * This is typically of the form: * [0.5, 0.5] or some other probability distribution summing to one */ public INDArray output(INDArray input, boolean train) { WorkspaceMode cMode = layerWiseConfigurations.getTrainingWorkspaceMode(); layerWiseConfigurations.setTrainingWorkspaceMode(layerWiseConfigurations.getInferenceWorkspaceMode()); MemoryWorkspace workspace = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal); try (MemoryWorkspace wsE = workspace.notifyScopeEntered()) { INDArray ret = silentOutput(input, train).detach(); layerWiseConfigurations.setTrainingWorkspaceMode(cMode); return ret; } }
layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( ComputationGraph.workspaceConfigurationExternal, ComputationGraph.workspaceExternal); MemoryWorkspace cache = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( ComputationGraph.workspaceConfigurationCache,
MemoryWorkspace workspace = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SINGLE ? Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceExternal) : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationFeedForward, workspaceFeedForward); MemoryWorkspace pretrain = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SINGLE ? Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceExternal) : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(
layer.conf().setPretrain(true); MemoryWorkspace workspace = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SINGLE ? Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceExternal) : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationFeedForward, workspaceFeedForward); MemoryWorkspace pretrain = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SINGLE ? Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceExternal) : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(
layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SINGLE ? Nd4j.getWorkspaceManager() .getWorkspaceForCurrentThread(workspaceExternal) if (layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.SEPARATE) { Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread(workspaceFeedForward).initializeWorkspace();
layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal);
if (iterator.asyncSupported()) { iter = new AsyncDataSetIterator(iterator, Math.min(Nd4j.getAffinityManager().getNumberOfDevices() * 2, 2), layerWiseConfigurations.getTrainingWorkspaceMode() != WorkspaceMode.NONE); destructable = true; } else { layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal); MemoryWorkspace cache = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( ComputationGraph.workspaceConfigurationCache,
layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationTBPTT, workspaceTBPTT); MemoryWorkspace workspace = layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal); if (layerWiseConfigurations.getTrainingWorkspaceMode() != WorkspaceMode.NONE) { workspace.initializeWorkspace(); workspaceT.initializeWorkspace();
WorkspaceMode cMode = layerWiseConfigurations.getTrainingWorkspaceMode(); layerWiseConfigurations.setTrainingWorkspaceMode(layerWiseConfigurations.getInferenceWorkspaceMode()); layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal);
layerWiseConfigurations.getTrainingWorkspaceMode(), layerWiseConfigurations.getInferenceWorkspaceMode());
layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( workspaceConfigurationExternal, workspaceExternal); layerWiseConfigurations.getTrainingWorkspaceMode() == WorkspaceMode.NONE ? new DummyWorkspace() : Nd4j.getWorkspaceManager().getWorkspaceForCurrentThread( ComputationGraph.workspaceConfigurationCache,