how to use TensorFlow?

---

what is TensorFlow

Tensors, in general, are simply arrays of numbers, or functions, that transform according to certain rules under a change of coordinates. TensorFlow is an open source software library for doing graph-based computations quickly. It does this by utilizing the GPU(Graphics Processing Unit), and also making it easy to distribute the work across multiple GPUs and computers.

Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them. This model is used for learning vector representations of words, called word embeddings. One of the tasks at which it excels is implementing and training deep neural networks.

What can TensorFlow do

Google’s official blog gives quite a bit of information. TensorFlow - Google’s latest machine learning system, open sourced for everyone.

The short answer: it allows machine learning researcher to experiment with deep neural nets more easily. Here, “more easily” means it takes less time to construct artificial neural networks, less time to train them (this is arguable, as is always the case when speed is concerned), and less time to deploy them.

Do not expect: do not expect magic things to come from TensorFlow. There is none. A wealth of deep learning frameworks already exist in the wild and TensorFlow is just one of them. It will not give you powerful AI overnight, nor will it help your smartphone to recognize cats. It is a tool for deep learning research that will hopefully make it a bit less cumbersome.

Who should be excited: deep learning researchers who construct and experiment with large number of networks, researchers who are too lazy to get their hands dirty on serious code, and companies that benefit from being able to quickly train and deploy their models. Basically if you consider yourself a member of the general public, it’s got nothing to do with you.

start learning

The TensorFlow website has quite a bit of documentation and tutorials, but these tend to assume some (or a lot) of prior machine learning and artificial intelligence knowledge.

Beyond knowing ML and AI, you are also expected to be very comfortable with the Python programming language.

For this reason, starting to learn how to use TensorFlow might actually mean first learning more Python, rather than anything directly related to Machine Learning.

Assuming you know your Python, but not machine learning, then you have some catching up to do. You could check out this Practical Machine Learning with Python tutorial, which covers concepts and algorithms related to machine learning, giving theory, application, and a break down of the inner-workings of each of them.

If you already know both Python and Machine learning basics, but you do not yet know Deep Learning/TensorFlow, then you can start at the Introduction to Neural Networks part.

If you already know neural networks/deep learning, then either start with the Installing TensorFlow tutorial, or you can begin with the TensorFlow basics tutorial, which leads immediately into actually modeling a deep neural network.