General Overview of TensorFlow

TensorFlow is an open source framework from Google for scientific and numerical computation based on dataflow graphs that stand for the TensorFlow's execution model. The dataflow graphs used in TensorFlow help the machine learning experts to perform more advanced and intensive training on the data for developing deep learning and predictive analytics models. In 2015, Google open sourced the TensorFlow and all of its reference implementation and made all the source code available on GitHub under the Apache 2.0 license. Since then, TensorFlow has achieved wide adoption from academia and research to the industry, and following that recently the most stable version 1.x has been released with a unified API.

As the name TensorFlow implies, operations are performed by neural networks on multidimensional data arrays (aka flow of tensors). This way, TensorFlow provides some widely used and robust implementation linear models and deep learning algorithms.

Deploying a predictive or general purpose model using TensorFlow is pretty straightforward. The thing is that once you have constructed your neural networks model after necessary feature engineering, you can simply perform the training interactively using plotting or TensorBoard (we will see more on it in upcoming sections). Finally, you deploy it eventually after evaluating it by feeding it some test data.

Since we are talking about the dataflow graphs, nodes in a flow graph correspond to the mathematical operations, such as addition, multiplication, matrix factorization, and so on, whereas, edges correspond to tensors that ensure communication between edges and nodes, that is dataflow and controlflow.

You can perform the numerical computation on a CPU. Nevertheless, using TensorFlow, it is also possible to distribute the training across multiple devices on the same system and train on them, especially if you have more than one GPU on your system so that these can share the computational load. But the precondition is if TensorFlow can access these devices, it will automatically distribute the computations to the multiple devices via a greedy process. But TensorFlow also allows the program, to specify which operations will be on which devices via name scope placement.

The APIs in TensorFlow 1.x have changed in ways that are not all backward compatible. That is, TensorFlow programs that worked on TensorFlow 0.x won't necessarily work on TensorFlow 1.x.

The main features offered by the latest release of TensorFlow are:

Throughout the next lesson, we will see how to achieve these features for predictive analytics.