This is going to be very common use case where deep learning classification task can be greatly improved by providing extra meta context, such as person’s age, or image location, or user’s IP address - in addition to main data, such as image or user session activity or tweet.
While we touched multi-input at Lesson 7 and Keras has couple of (incomplete) basic examples of multi-input - I couldn’t find anywhere a complete, simple, working piece of code leveraging model.fit_generator() where input comes in batches.