Zhang Yuhong an excerpt from the "depth study of the United States", as their study notes!
Why TensorFlow to take such a name? Naturally, this is a luxurious.
Tensor Flow was named for its operation principle of "Tensor" was intended to "tensor", "tensor" generally refers to a multi-dimensional matrix. In depth study programs, mostly higher than the two-dimensional data, so take advantage of the core features of the data of the depth of the learning process to name, it makes sense. "Flow" is meant the "flow", it means that the data flow is calculated based on FIG. Together, "TensorFlow" means that, tensor flow from one end to the data flow graph calculation process at the other end. It vividly described flow, transmission, analysis and processing of complex data structures in the model of artificial neural networks (see Figure 11-3).
Towel machine learning, the value typically composed of four types:
- A scalar (scalar): is a numeric value that is a minimum calculation unit, such as "1" or "3.2" and the like.
- To iil (vector): one-dimensional array constituted by a number of scalars, such as [1, 3.2, 4.6] and the like.
- Matrix (matrix): two-dimensional array consisting of a scalar.
- Tensor (tensor): Qu multidimensional (generally n> 3) composed of a set of data array. Understood as high-dimensional matrix.
Tensor Flow is an excellent deep learning framework, which all have excellent performance in many aspects. For example, the code design is very simple neural network architecture, deployment is also more convenient. Especially those with strong technical strength of Google for its "Platform", who owned numerous crooked, to a large extent to ensure that its community activity, which also led to TensorFlow iterative update very fast evolution of technology on the road, basically every Zhou has submitted thousands of lines of code.
Tensor Flow advantage is mainly in the following three aspects:
(1) TensorFlow has a very intuitive framework. As the name suggests, it has a "tensor flow." Users can easily, visually see every aspect of the tensor flow (needs the TensorBoard, in later chapters will be mentioned).
(2) TensorF! Ow can easily be on the CPU / GPU deployment, distributed computing, large data provide support for the computing capability analysis.
(3) TensorF! Ow cross-platform, and strong flexibility. TensorFlow not only run on Linux, Mac and Windows systems, you can even work in a mobile terminal.
Of course, TensorFlow also has shortcomings. Mainly in its underlying code comparison, the user needs to write a lot of code, and there are many similar features, users had to "reinventing the wheel." But have fundamental, TensorFlow or with strong technical accumulation, stable performance, a ride Red, "windy" at the top of many deep learning framework.