The overall algorithm (HumanAR.scala) has the following workflow:
- Loading the data
- Defining hyperparameters
- Setting up the LSTM model using imperative programming and the hyperparameters
- Applying batch wise training, that is, picking batch size data, feeding it to the model, then at some iterations evaluating the model and printing the batch loss and the accuracy
- Output the chart for the training and test errors
The preceding steps can be followed and constructed by way of a pipeline:
Figure 10: MXNet pre-built binary generated
Now let's start the implementation step-by-step. Make sure that you understand ...
