Shulou(Shulou.com)06/01 Report--

Google open source TensorFlow reinforcement learning framework example analysis, many novices are not very clear about this, in order to help you solve this problem, the following editor will explain in detail for you, people with this need can come to learn, I hope you can get something.

Google announced Dopamine, an open source reinforcement learning framework based on TensorFlow.

Reinforcement learning is an artificial intelligence (AI) technology that uses rewards (or punishments) to drive agent toward specific goals, such as the previously popular Alpha Go defeating the top human go players and the Open AI Five against human professional players in Dota2. At the same time, reinforcement learning is also a core part of DeepMind's Deep Q Network (DQN), which can be distributed among multiple workers, for example, to achieve "superman" performance in Atari 2600 games. The trouble is that reinforcement frameworks take time to master a goal, which is often inflexible and unstable.

That's why Google has come up with an alternative: Dopamine, an open source reinforcement learning framework based on TensorFlow, which is available from Github starting today.

(https://github.com/google/dopamine/tree/master/docs#downloads)

Google researchers say their open source TensorFlow reinforcement learning framework emphasizes three points: flexibility, stability and repeatability.

Inspired by rewarding motivational behavior in the brain, one of the main components, and reflecting the link between neuroscience and reinforcement learning, this platform aims to enable speculative research to drive fundamental discoveries, and this version also includes a set of colabs that illustrates how to use the entire framework.

Ease of use

To this end, it includes a set of well-written code (15 Python files) that focuses on the Arcade learning environment (a platform for evaluating AI technology with video games) and four different machine learning models: the Deep Q Network (DQN) mentioned above; C51; a simplified version of Rainbow agent; and Implicit Quantile Network agent. Clarity and simplicity are two key considerations in the design of this framework.

Repeatability

In order to achieve repeatability of reinforcement learning, the code provides complete test coverage and training data (in JSON and Python pickle formats) in 60 games supported by the Arcade learning environment, and follows the best practices of standardizing results for empirical evaluation.

Benchmark test

It is important for new researchers to benchmark their ideas quickly. Google provides complete training data for four agents, including 60 games supported by ALE, in the format of Python pickle files (for agents trained using Google framework) and JSON data files (used to compare agents trained with other frameworks). Google also provides a website that researchers can use to quickly visualize the training runs of all providing agents in all 60 games.

Google's four agents are trained to run on Seaquest (Seaquest is Atari supported by ALE

In addition, Google has launched a website that allows developers to quickly visualize the operation of multiple agents in training. It also provides trained models, raw statistical logs and TensorFlow event files for TensorBoard dynamic graph drawing, TensorBoard is a web application visualization suite.

"our hope is that the flexibility and ease of use of our framework will enable researchers to try new ideas, whether gradual or radical."

Is it helpful for you to read the above content? If you want to know more about the relevant knowledge or read more related articles, please follow the industry information channel, thank you for your support.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.