Machine learning libraries

I am write an algo using machine learning. It seems that Keras is not installed on the play server, since the modified version of the starter algo that imports Keras doesn’t work. Are there any available machine learning libraries on the play server?

In our current alpha state, we cannot guarantee the availability of any packages, though we do have plans to include some later on. However, with the new Starter Kit, you will have the ability to run games locally, so you could train two machine learning bots against one another and then release them into the arena. If this doesn’t work for you, you may have to find a creative solution or change plans for now.

Good luck in the Arena!
-C1 Ryan

Hello C1Ryan,

we cannot guarantee the availability of any packages, though we do have plans to include some later on.

Are there any updates regarding the availability of packages? I think for starters that NumPy would be sufficient :slight_smile:

-Jakob

1 Like

It’s a complicated issue so it may take some time to figure it out. Especially since we will be releasing java and other language support soon so we will need a multi-language solution.

But for now you could copy the library files into your algo folder and import those files into your code directly.

Can I ask how did you do to automatically run your local engine to judge your model? (if you did)

me? I did nothing so far…

But for now you could copy the library files into your algo folder and import those files into your code directly.

Ok got that, this is working lokaly but now the upload of the zip folder fails with “Your algo failed to upload. Please upload the correct Zipped Asset.”
… Is there a maximum upload size? The ziped starter_algo with numpy in it has 15 MB. The ziped starter_algo without uploads without trubble ~300kb

I initially tried to make a genetic algorithm that ran the local game engine to determine the fitness of each specimen. From inside a python program, you can import the ‘subprocess’ library and use it to execute anything the windows command line or unix terminal could run, including the local game engine with the appropriate arguments passed.
https://docs.python.org/3.6/library/subprocess.html

However, I found that matches in the local engine, at least in the unmodified form I was using, take far too long to use as feedback for machine learning.

A possible solution by @C1Junaid :

Thank you for the tip!

I had the same idea as you and came to the same conclusion haha.
I think I will still try it just for the fun.

1 Like