This article compares the TensorFlow and paddlepaddles framework mainly from the five aspects of framework overview, system architecture, programming model, distributed architecture and framework comparison.

As the deep learning framework developed by the two major international search engines, it has different focuses, but it also provides elegant and concise design architecture, and it is still developing.

For paddlepaddles, its ease of use and native and fast business integration are a very favorable weapon for the domestic Internet companies that are winning with speed. The flexibility and relatively scientific nature of TensorFlow is a big advantage in the field of AI research.

An overview of the framework

PaddlePaddle was developed in 2013 and has been accompanied by the rapid growth of baidu's training data in advertising, text, image and voice, as well as the algorithmic requirements in baidu takeout, search and unmanned driving.

Based on the single GPU training platform, baidu Deep Learning lab has developed the multi-machine Parallel GPU training platform for paddles and Parallel Distributed Deep Learning.

Since paddlepaddles was made open source, its design and paddlepaddles have been focused on being "easy to use, efficient, flexible and extensible".

Just as the design positioning of the official website: An easy-to-use, Efficient, Flexible and Scalable Deep Learning Platform. Below are the Paddle's official website:

The paddlepaddles framework was made open source in September 2016 and its biggest feature and paddlepaddles are easy to use.

Therefore, many algorithms are completely encapsulated, not only for the existing CV, NLP and other algorithms (such as VGG, ResNet, LSTM, GRU, etc.).

It is available under the models model library (https://www.fanke123.com) module, Encapsulates the word vector (including Hsigmoid acceleration vector training and noise estimation word vector training), the language of the RNN model, clickthrough rate forecast, text classification, sorting, learning, information retrieval and search engine research one of the core problems), structured semantic model, named entity recognition, sequence to sequence learning, reading comprehension, automatic question answering, image classification, target detection, scene text recognition, speech recognition, and other technical areas of artificial intelligence and general solution.

Each of the above solutions is designed for a technical scenario, so the developer may only need to have a slight understanding of the source code principles, follow the example on the official website to run the commands, change their own data, and modify some super parameters to run.

Its second major feature is distributed deployment, and it is currently the only deep learning library that supports Kubernetes. This is further explained in the "distributed architecture" section of this article.

The description of TensorFlow on the official website of TensorFlow is An open-source software library for Machine Intelligence, An open source Machine learning library.

TensorFlow is currently the most popular ai algorithm engine in terms of number of users and activity.

So what are the paddlepaddles advantages for developers?

First, ease of use.

In addition, PaddlePadddle is the open source framework of domestic giant baidu. Its native feature is not only in line with the usage habits of Chinese people, but also attaches great importance to the use scenarios and solutions of mainstream Internet technology.

Second, faster speed. As mentioned above, paddlepaddles' code and design are much simpler, and using them for model development can obviously save developers some time. This makes paddlepaddles suitable for industrial applications, especially for scenarios requiring rapid development.

Source: royal international /www.fanke123.com