TensorBoard

The complexity of a computation graph gets high even for moderately sized problems. Large computational graphs that represent complex machine learning models can become quite confusing and hard to understand. Visualization helps in easy understanding and interpretation of computation graphs, and thus accelerates the debugging and optimizations of TensorFlow programs. TensorFlow comes with a built-in tool that allows us to visualize computation graphs, namely, TensorBoard.

TensorBoard visualizes computation graph structure, provides statistical analysis and plots the values captured as summaries during the execution of computation graphs. Let's see how it works in practice.