Once Tensorflow was open source, it immediately attracted a large number of developers to follow up. Tensorflow widely supports many functions, including image, handwriting, speech recognition, prediction and natural language processing. TensorFlow follows the Apache 2.0 open source protocol.
TensorFlow released its version 1.0 on February 5, 20 17, which is an integration of the previous eight imperfect versions. Here are some reasons for TensorFlow's success:
TensorFLow provides these tools:
TensorBroad is a well-designed visual network construction and display tool.
TensorFlow Serving can easily configure new algorithms and environments by maintaining the same server architecture and API. TensorFlow service also provides out-of-the-box models that can be easily extended to support other models and data.
The alpha version also supports TensorFlow programming interfaces including Python and C++, Java, Go, R and Haskell languages. In addition, TensorFlow also supports the cloud environments of Google and Amazon.
Tensorflow version 0.12 supports Windows 7, 8 and Server 20 16 systems. Because of the C++ feature library, TensorFlow class library can be compiled and optimized on the ARM architecture platform. This means that you can deploy trained models on various servers and mobile devices without additional implementation of model decoders or Python interpreters.
TensorFlow provides a detailed network layer so that users can build new and complex layer structures without having to implement them from the bottom. A subgraph allows users to view and restore data on any edge of the graph. This is very useful for debugging complex calculations.
Distributed TensorFlow was introduced in version 0.8, which provides parallel computing support and allows different parts of the model to be trained in parallel on different devices.
TensorFlow has taught at Stanford University, Berkeley College, University of Toronto and Uda City (an online school established in March 2065438+2006).
TensorFlow has the following disadvantages:
Each calculation process must be constructed as a graph without symbolic cycle, which makes some calculations difficult;
Without 3-D convolution, video recognition is impossible;
Even though it is 58 times faster than the original version (0.5), its execution performance is still not as good as that of competitors.