It is essential in machine learning engineering to know how to proceed during the development of a system to avoid pitfalls and address common issues. The easiest way to create a machine learning system, that saves you money and time, is to reuse code and pretrained models that have been applied to similar problems to your own. If this does not cover your needs, then you may need to train your own CNN architecture as this can sometimes be the best way to solve your problem. However, one of the biggest challenges to face is finding large scale, publicly available datasets that are tailor-made to your problem. Therefore, it is often the case that you may need to create your own dataset. When creating your own dataset it is very crucial to organize it appropriately in order to insure successful model training.

In this chapter we will present and discuss the day-to-day workflow that will help you to answer the following questions:

• How should I split my dataset?
• Is my dataset representative enough of my problem?
• How complex should my model be to be both efficient and accurate?
• What is the best method to evaluate my model?
• How should I structure my code?