First released in 2015, Keras was designed as an interface to enable fast experimentation with neural networks. As such, it relied on TensorFlow or Theano (another deep learning framework, now deprecated) to run deep learning operations. Known for its user-friendliness, it was the library of choice for beginners.
Since 2017, TensorFlow has integrated Keras fully, meaning that you can use it without installing anything other than TensorFlow. Throughout this book, we will rely on tf.keras instead of the standalone version of Keras. There are a few minor differences between the two versions, such as compatibility with TensorFlow's other modules and the way models are saved. For this reason, readers must make sure to use the correct version, as follows:
- In your code, import tf.keras and not keras.
- Go through the tf.keras documentation on TensorFlow's website and not the keras.io documentation.
- When using external Keras libraries, make sure they are compatible...
