Implementing and training a model in TF
While PyTorch is oriented towards research projects, TF puts more emphasis on industry use cases. While the deployment features of PyTorch, Torch Serve, and Torch Mobile are still in the experimental phase, the deployment features of TF, TF Serve, and TF Lite are stable and actively in use. The first version of TF was introduced by the Google Brain team in 2011 and they have been continuously updating TF to make it more flexible, user-friendly, and efficient. The key difference between TF and PyTorch was initially much larger, as the first version of TF used static graphs. However, this situation has changed with version 2, as it introduces eager execution, mimicking dynamic graphs known from PyTorch. TF version 2 is often used with Keras, an interface for ANN (https://keras.io). Keras allows users to quickly develop DL models and run experiments. In the following sections, we will walk you through the key components of TF.
