ND4JWorkspaceException-dl4j

gistfile1.txt

Error Log:
java.lang.IllegalStateException: Backprop: array (ACTIVATION_GRAD) workspace validation failed (vertex lstm1 - class: GravesLSTM) - array is defined in incorrect workspace
	at org.deeplearning4j.nn.graph.ComputationGraph.validateArrayWorkspaces(ComputationGraph.java:1708)
	at org.deeplearning4j.nn.graph.ComputationGraph.calcBackpropGradients(ComputationGraph.java:2435)
	at org.deeplearning4j.nn.graph.ComputationGraph.computeGradientAndScore(ComputationGraph.java:1319)
	at org.deeplearning4j.nn.graph.ComputationGraph.computeGradientAndScore(ComputationGraph.java:1280)
	at org.deeplearning4j.optimize.solvers.BaseOptimizer.gradientAndScore(BaseOptimizer.java:178)
	at org.deeplearning4j.optimize.solvers.StochasticGradientDescent.optimize(StochasticGradientDescent.java:60)
	at org.deeplearning4j.optimize.Solver.optimize(Solver.java:54)
	at org.deeplearning4j.nn.graph.ComputationGraph.fit(ComputationGraph.java:1104)
	at org.deeplearning4j.nn.graph.ComputationGraph.fit(ComputationGraph.java:1050)
	at org.deeplearning4j.nn.graph.ComputationGraph.fit(ComputationGraph.java:944)
	at com.liweigu.dls.competition.srad.RadLSTM.run(RadLSTM.java:102)
	at com.liweigu.dls.competition.srad.RadLSTMTest.testRun(RadLSTMTest.java:10)
Caused by: org.nd4j.linalg.workspace.ND4JWorkspaceException: Array workspace validation failed: Array of type ACTIVATION_GRAD should be in workspace "WS_LAYER_ACT_1" but is in workspace "WS_LAYER_WORKING_MEM"
	at org.nd4j.linalg.workspace.BaseWorkspaceMgr.validateArrayLocation(BaseWorkspaceMgr.java:221)
	at org.deeplearning4j.nn.workspace.LayerWorkspaceMgr.validateArrayLocation(LayerWorkspaceMgr.java:66)
	at org.deeplearning4j.nn.graph.ComputationGraph.validateArrayWorkspaces(ComputationGraph.java:1699)
	... 33 more
  
Key Code:
		int cnn1Stride = 4;
		int lstmHiddenCount = 100;
		LOGGER.info("lstmHiddenCount = " + lstmHiddenCount);
		int channels = 1;
		int kernelSize = 3; // 5
		Map<String, InputPreProcessor> inputPreProcessors = new HashMap<String, InputPreProcessor>();
		inputPreProcessors.put("cnn1", new RnnToCnnPreProcessor(501, 501, channels));
		inputPreProcessors.put("lstm1", new CnnToRnnPreProcessor(120, 120, channels)); // 120 = Math.pow(14400, 0.5)
		GraphBuilder graphBuilder = builder.graphBuilder().pretrain(false).backprop(true).backpropType(BackpropType.Standard)
	            .addInputs("inputs")
	            .addLayer("cnn1", new ConvolutionLayer.Builder(new int[] { kernelSize, kernelSize }, new int[] { cnn1Stride, cnn1Stride }, new int[] { 0, 0 })
	    				.nIn(channels).nOut(50).activation(Activation.RELU).weightInit(WeightInit.RELU).build(), "inputs")
	            // Output: (501 - kernelSize + 0) / cnn1Stride + 1 = 125 --> 125 * 125 * 50 = 781250
				.addLayer("cnn2", new SubsamplingLayer.Builder(new int[] { 2, 2 }, new int[] { 2, 2 }).build(), "cnn1")
				// (125-2+0)/2+1 = 62
				.addLayer("cnn3",
						new ConvolutionLayer.Builder(new int[] { kernelSize, kernelSize }, new int[] { 5, 5 }, new int[] { 0, 0 })
						.nIn(50).nOut(100).activation(Activation.RELU).weightInit(WeightInit.RELU).build(), "cnn2")
				// Output: (62 - kernelSize + 0) / 5 + 1 = 12 --> 12 * 12 * 100 = 14400
				.addLayer("lstm1", new GravesLSTM.Builder().activation(Activation.SOFTSIGN)
						.nIn(14400).nOut(lstmHiddenCount).build(), "cnn3")
	            .addLayer("lstm2", new GravesLSTM.Builder().activation(Activation.SOFTSIGN)
	            		.nIn(lstmHiddenCount).nOut(lstmHiddenCount).build(), "lstm1")
	            .addVertex("thoughtVector", new LastTimeStepVertex("inputs"), "lstm2")
	            .addLayer("output", new OutputLayer
	            		.Builder(LossFunctions.LossFunction.MSE) // MSE, MEAN_ABSOLUTE_ERROR
	    				.activation(Activation.IDENTITY)
	    				.nIn(lstmHiddenCount)
	    				.nOut(501 * 501).build(), "thoughtVector")
				.setOutputs("output");
		graphBuilder.setInputPreProcessors(inputPreProcessors);

		ComputationGraph multiLayerNetwork = new ComputationGraph(graphBuilder.build());
		multiLayerNetwork.init();