API¶
The main entry point is in rnn.
This will initialize an instance of Engine
which has the high level logic to iterate through epochs.
For each GPU (or CPU), an instance of Device
will be created which does all the calculation and the model update.
The model is an instance of Network.LayerNetwork
and consists of multiple hidden layers of class NetworkBaseLayer.Layer
and one or more output layers of class NetworkOutputLayer.OutputLayer.
The model update code, i.e. the optimization methods such as SGD
are implemented in Updater.
ActivationFunctionsBestPathDecoderBundleFileCTCCachedDatasetCachedDataset2ConfigCustomLSTMFunctionsDatasetDebugDebugHelpersDeviceEngineEngineBaseEngineBatchEngineTaskEngineUtilExternalFsaFunctionLoaderGeneratingDatasetHDFDatasetHyperParamTuningInvLearningRateControlLmDatasetLogMetaDatasetMultiBatchBeamNativeOpNetworkNetworkBaseLayerNetworkCNNLayerNetworkCopyUtilsNetworkCtcLayerNetworkDescriptionNetworkHiddenLayerNetworkLayerNetworkLstmLayerNetworkOutputLayerNetworkRecurrentLayerNetworkStreamNetworkTwoDLayerNormalizationDataNumpyDumpDatasetOpBLSTMOpInvAlignOpLSTMOpLSTMCellOpLSTMCustomOpLSTMRecOpNumpyAlignPretrainRawWavDatasetRecurrentTransformServerSprintCacheSprintControlSprintDatasetSprintErrorSignalsSprintExternInterfaceSprintInterfaceStereoDatasetTFDataPipelineTFEngineTFKenLMTFNativeOpTFNetworkTFNetworkLayerTFNetworkNeuralTransducerTFNetworkRecLayerTFNetworkSegModLayerTFNetworkSigProcLayerTFSprintTFUpdaterTFUtilTaskSystemTaskSystem_exampleTheanoUtilTorchWrapperTwoStateBestPathDecoderTwoStateHMMOpUpdaterUtilbetter_exchookrnnsetup