Emojify_V2 Keras issues

Course Q&A Deep Learning Specialization Sequence Models
1

Still struggling to figure out how to put together a Keras pipeline.

We have:
# Propagate sentence_indices through your embedding layer
# (See additional hints in the instructions).

Except I don’t see any relevant additional hints. Is this supposed to create a layer? Should this be where we create the input layer? I have no idea what to do here, except I am setting a variable called ‘embeddings’ that isn’t obvious where it is used.

I do see:
The embedding_layer that is returned by pretrained_embedding_layer is a layer object that can be called as a function, passing in a single argument (sentence indices). So I assume I have to do that at some point, but that gives me a 4x2 embeddings tensor that isn’t passable to my LSTM layer.

Then we create a LSTM layer. The instructions are passing ‘processed_inputs’ which makes me think I should create an input layer and do something with it. But I already have an embedding_layer created, which looks like it should be the first layer. So I have no idea what gets passed to the first LSTM or what gets created before that.

The rest of the pipeline is starightforward until you get to the ‘Model’ line, which the instructions tell us nothing about except it takes ‘input’ and ‘output’ either of which could be many different things in this function.

2

In general, there are two ways to specify the input.
One is via an input layer.
The other is by using an input argument on the first layer you define.

3

Do you mean that:
X = Input(shape=input_shape, dtype=‘int32’)(data)
X = SomeLayer(‘spam’, ‘eggs’)(X)

is equivalent to:
X = SomeLayer(‘spam’, ‘eggs’)(data)

If that is what you are trying to say, it is helpful to know, but doesn’t really say much about what to pass to my first LTSM layer or model functions.