TensorFlow Tutorial - Layers API
Notes by Magnus Erik Hvass Pedersen: https://github.com/Hvass-Labs/TensorFlow-Tutorials/blob/master/03B_Layer...
It is important to use a builder API when constructing Neural Networks in TensorFlow because it makes it easier to implement and modify the source-code. This also lowers the risk of bugs.
Many of the other tutorials used the TensorFlow builder API called PrettyTensor for easy construction of Neural Networks. But there are several other builder APIs available for TensorFlow. PrettyTensor was used in these tutorials, because at the time in mid-2016, PrettyTensor was the most complete and polished builder API available for TensorFlow. But PrettyTensor is only developed by a single person working at Google and although it has some unique and elegant features, it is possible that it may become deprecated in the future.
This tutorial is about a small builder API that has recently been added to TensorFlow version 1.1. It is simply called Layers or theLayers API or by its Python name tf.layers. This builder API is automatically installed as part of TensorFlow, so you no longer have to install a separate Python package as was needed with PrettyTensor.
This tutorial is very similar to Tutorial #03 on PrettyTensor and shows how to implement the same Convolutional Neural Network using the Layers API. It is recommended that you are familiar with Tutorial #02 on Convolutional Neural Networks.