- What is Keras compared to TensorFlow? What is its purpose?
Keras was designed as a wrapper around other deep learning libraries to make development easier. TensorFlow is now fully integrated with Keras through tf.keras. It is best practice to use this module to create models in TensorFlow 2.
- Why does TensorFlow use graphs? How can they be created manually?
TensorFlow relies on graphs to ensure model performance and portability. In TensorFlow 2, the best way to create graphs manually is to employ the tf.function decorator.
- What is the difference between eager execution mode and lazy execution mode?
In lazy execution mode, no computation is performed until the user specifically asks for a result. In eager execution mode, every operation is run when it is defined. While the former can be faster thanks to graph optimizations, the latter is easier to use and easier to debug. In TensorFlow 2, lazy execution mode has been deprecated in favor of eager execution mode.
- How do you...
