We have just mentioned that building a suitable expression database is a lot of work. To be able to build an end-to-end working expression analysis solution (and fit it all into a single chapter of this book), we will take some liberties with our project:

• We will limit our model's ability to detect and classify only two emotions—happiness and sadness
• We will supply only a limited amount of expression data to train our model

Obviously, in the real world, our second assumption is a risky one; as in any ML model, less training data typically produces a less valuable result.