TensorFlow is one of the most popular libraries for working with neural networks. In the preceding section, we saw how we can use it as the backend engine of Keras. It is an open source, high-performance library that can actually be used for any numerical computation. If we look at the stack, we can see that we can write TensorFlow code in a high-level language such as Python or C++, which gets interpreted by the TensorFlow distributed execution engine. This makes it quite useful for and popular with developers.

The way TensorFlow works is that you create a Directed Graph (DG) to represent your computation. Connecting the nodes are the edges, the input, and the output of the mathematical operations. Also, they represent arrays of data.