Developing classification and prediction models using deep learning networks involves extensive experimentation to arrive at models with high-quality performance. To help with this process, there are various methods that are very useful for visualizing and controlling network training. In this chapter, we went over four such useful methods. We saw that TensorBoard provides a tool that we can use to assess and compare model performance after training the network with different architectures and other changes in the model. The advantage of using TensorBoard lies in the fact that it brings all the necessary information together in one place in a user-friendly way. There are also situations where we want to understand how the main features or variables on a specific prediction are influenced when using a classification or prediction model. In such situations, we can visualize the impact that the main features will have using LIME.

Another useful tip that we illustrated in this chapter is visualization with the help of tfruns. When developing a deep network model, we come across various plots and summaries related to a specific model. Using tfruns, we can visualize all the information in one place with the help of an interactive screen. Another tip or trick that will be very useful in the journey ahead is the use of callbacks to automatically stop the training process when a suitable classification or prediction model has been developed. All the methods that were discussed in this chapter can be very useful for the journey ahead, especially when you're working on complex and challenging problems.