This week

• [x] Implement the custom training step
• [x] Explore feature embedding and encoding types

Advanced: Customized Training

The past week training procedure is simple, but does not give you much control.

So now let's unpack the training loop, and implement it ourselves. This gives a starting point if, for example, to implement curriculum learning to help stabilize the model's open-loop output.

What is curriculum learning?

<aside> 💡 Curriculum learning describes a type of learning in which you first start out with only easy examples of a task and then gradually increase the task difficulty. We humans have been learning according to this principle for decades, yet we don’t transfer it to neural networks and instead let them train on the whole data set with all its difficulties from the beginning on. Link

</aside>

We will use tf.GradientTape to track the gradients. You can learn more about this approach by reading the eager execution guide.

The procedure works as follows:

• First, initialize the RNN state. We do this by calling the tf.keras.Model.reset_states method.
• Next, iterate over the dataset (batch by batch) and calculate the predictions associated with each.
• Open a tf.GradientTape, and calculate the predictions and loss in that context.
• Calculate the gradients of the loss with respect to the model variables using the tf.GradientTape.grads method.
• Finally, take a step downwards by using the optimizer's tf.train.Optimizer.apply_gradients method.

Text generation with an RNN | TensorFlow Core

Le't create again our model as last week, the model is the same:

# Build the model
model = build_model(
  vocab_size = len(vocab),
  embedding_dim=embedding_dim,
  rnn_units=rnn_units,
  batch_size=BATCH_SIZE)

Now we set the optimizer for our parameters