[PYTHON] Reinforcement learning 28 colaboratory + OpenAI + chainerRL

(The chokozainer RL was updated on December 8, 2019.)

It is intended for AI beginners from junior high school students to university students. I have written up to 27 reinforcement learning series. It was about one month because it was one pace a day. I will write a summary that makes it easier to start from here. There is nothing new.

If you want to start machine learning using GPU easily, colaboratory is free, so I recommend it. No need for troublesome installation. However, that alone will not work, so let's summarize the procedure. I chose chainerRL as the framework. I like tensorflow, but I haven't used it, so ... I think I'll try using tensorflow soon. I will write it in the middle, but let's look at the source code as much as possible. It is published on github. The function name is easy to understand, so I think it's easy to understand. Chainer is easy to read in English, probably because it is made by Japanese people. Or rather, if you translate it into Japanese with chrome, it will be in proper Japanese. What about tensorflow? ?? ?? is. We have released chokozainerRL, which is a wrapper for chainerRL. I haven't done much, but I hope it will be useful for "human learning" in reinforcement learning.

1 Get a Google account

Please create from here.

Create a google account https://support.google.com/accounts/answer/27441?hl=ja

2 Open the Colaboratory page

Open from here. https://colab.research.google.com/notebooks/welcome.ipynb?hl=ja Bookmark it.

3 Open the chokozainer sample page and save it in your own folder

3-1 Open the notebook

File-Open Notebook

Select the GitHub tab. Search with chokozainer, ipynbs/abc.ipynb Choose.

The opened abc.ipynb cannot be used as it is, so make a copy on the drive.

Then rename the copied file.

4 Run the notebook.

Running a notebook is explained in detail on various sites, so please do it yourself. Before learning, you could only do a few steps, but after learning, you can see that you can balance up to 200 steps of setting. You can make a video like this.

The learning execution result looks like this.

Since elapsed is the execution time (seconds), learning will be completed in about 15 minutes.