Comparison with AlphaGo

According to DeepMind's paper, AlphaGo Zero was able to outperform AlphaGo in 36 hours, whereas the latter took months to train. In a head-to-head competition with the version of AlphaGo that defeated Lee Sedol, AlphaGo Zero won 100 games out of 100. What's significant about these results is that, even without initial human supervision, a Go playing program can reach superhuman-level proficiency more efficiently and is able to discover much of the knowledge and wisdom that humanity spent thousands of years and millions of games cultivating.

In the following sections, we will finally implement this powerful algorithm. Additional technical details of AlphaGo Zero will be covered as we go through the code.

Get Python Reinforcement Learning Projects now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.