I’m building a forward LSTM network, but I don’t know how to build the output layer. I connected a density layer to the LSTM layer, but it didn’t work. What’s wrong? The output layer as follow:
int adjBatchSize=batchSize*seqLenght; int inputSize=outputSizeOfLSTM; checkCublasErrors(cublasSgemm(culasHandle_, CUBLAS_OP_N, CUBLAS_OP_T, adjBatchSize, outputSize, inputSize, &FLOAT_ONE, dInput, adjBatchSize, dWeights, inputSize, &FLOAT_ZERO, dNetOutput, adjBatchSize)); cudaDeviceSynchronize();
dInput is the output of LSTM with the shape [seqLenght,batchSize,outputSizeOfLSTM] (CUDNN_RNN_DATA_LAYOUT_SEQ_MAJOR_PACKED); The shape of dNetOutput is [seqLenght,batchSize,outputSize]?