# Marrs Level of Analysis The **computational** level of the pyramid is the **why**. This is the level of objectives. The **algorithm** level is the **what**. What is the actual representation of the function we are learning. What is the structure that is going to answer that why? The **implementation** level is the **how**? How is that structure actually constructed in order to optimize the desired objective. How do you chose the what to answer the why?  These types of models can help us compartmentalize our thinking. It can help us to build [abstractions](Abstraction%20(Computer%20Science).md) so that we can solve the problems in isolation. This approach allows us to design optimization algorithms that are general and can be used with any loss function. --- Date: 20211124 Links to: [Machine-Learning](Machine-Learning.md) Tags: References: * [Machine Learning Basics](https://www.youtube.com/watch?v=zY2QgvPfSm8&list=PL_iWQOsE6TfVmKkQHucjPAoRtIJYt8a5A&index=6)