It is currently released under the Apache 2.0 open source license. TensorFlow programming interfaces includes Python and C++ with plans for Java, GO, R, and Haskell APIs. It is also supported in Google and Amazon cloud environment. TensorFlow is designed for large-scale distributed training and inference. The distributed Tensorflow architecture contains distributed master and worker services with kernel implementations. These include 200 standard operations, including mathematical, array manipulation, control flow, and state management operations written in C++. Unlike other DL libraries that are mainly focused on research (such as Theano) TensorFlow was designed for use both in research, development and production systems. It can run on single CPU systems, GPUs, mobile devices and large scale distributed systems of hundreds of nodes. In addition, TensorFlow Lite is TensorFlow lightweight solution for mobile and embedded devices [TensorflowLite]. It enables on-device ML inference with low latency and a small binary size but has coverage for a limited set of operators. It also supports hardware acceleration with the Android Neural Networks API.

Strong points

• By far the most popular DL tool, open source, fast involving, well-supported by the strong industrial company (Google).
• Powerful numerical library for dataflow programming that provides the basis for DL research and development.
• Efficiently works with mathematical expressions involving multi-dimensional arrays.
• Very well documented.
• GPU/CPU computing, mobile computing, high scalability of computation across machines and huge data sets.
• Higher layer of abstraction than Theano.

Weak points

• Still lower level API difficult to use directly for creating DL models.
• Every computational flow must be constructed as a static graph (although the Tensorflow Fold package tries to alleviate this problem), and lacks symbolic loops.

Return to Contemt