A good practice for productionalizing machine learning models is to separate the training and evaluation programs. Here, we illustrate an evaluation script expanded to include a unit test, model saving and loading, and evaluation.
For this recipe, we will show how to implement an evaluation script, using the above production criteria. The code actually consists of a training script and an evaluation script, but for this recipe, we will only show the evaluation script. As a reminder, both scripts can been seen in the online GitHub repository, https://github.com/nfmcclure/tensorflow_cookbook/.
For the example, we will implement the first RNN example from Chapter 9, Recurrent Neural Networks, which attempts to predict if a text message is spam or ham. We will assume the RNN model is trained and saved, along with the vocabulary.