The networks we have implemented so far were purely sequential, with a single path from inputs to predictions. The inception model differs from those, with its multiple parallel layers and branches. This gives us the opportunity to demonstrate that such operational graphs are not much more difficult to instantiate with the available APIs. In the following section, we will write an inception module using the Keras Functional API (refer to the documentation at https://keras.io/getting-started/sequential-model-guide/).
So far, we have mostly been using the Keras Sequential API, which is not well-adapted for multipath architectures (as its name implies). The Keras Functional API is closer to the TensorFlow paradigm, with Python variables for the layers being passed as parameters to the next ones to build a graph. The following code presents a simplistic model implemented with both APIs:
from keras.models import Sequential, Model
from keras...
