Rules Update

[Janzert]

The contest rules had not been updated for a few contests and the reality of what we want to enforce has changed a bit since then. The main functional change is the removal of the complete prohibition on file writing and OS threads. You can view the new rules on the Contest Rules page

[mac]

Actively using multiple processes or OS threads is not allowed. Some language runtimes will not operate without using multiple OS threads so we allow the creation of a limited number. But a bot found to be running active computation on more than one will be disqualified.

Question: using python, would it be allowed to launch a timer subprocess that could cut off the main computation thread just before time limit hits? Currently I have implemented it with a `while` loop and I am fine with it, but I would be interested in the answer for understanding the spirit of the rule better...

[Scryer]

The contest rules had not been updated for a few contests and the reality of what we want to enforce has changed a bit since then. The main functional change is the removal of the complete prohibition on file writing and OS threads. You can view the new rules on the Contest Rules page

Oh, cool. Better run a few tests to see if I can get CUDA running on the server graphics card. Should be an ideal problem for that! Oh, wait, did I say that out loud?

[jivinn]

Hi,

I was wondering if there's any way to create actual learning machines, so that feedback from each match could be used to update bot parameters. Now it's not possible, because created files aren't preserved between matches. My original plan was to make a really simple bot, which would then learn to fight like a spartan due time and experience, but at the moment I only have a fairly simple bot that sucks just as much after ten games as it does now..

Of course this learning can be done offline with matches between different generations, but I was hoping to be able to do it during the actual matches. Or is there another way to update some parts of the bot between matches?

Although, when I came up with my original plans, I thought that the bots would be fighting something like every ten minutes. This current frequency might not be enough for a simple learning algorithm.

[mac]

I was wondering if there's any way to create actual learning machines, so that feedback from each match could be used to update bot parameters. Now it's not possible, because created files aren't preserved between matches. My original plan was to make a really simple bot, which would then learn to fight like a spartan due time and experience, but at the moment I only have a fairly simple bot that sucks just as much after ten games as it does now.

Same situation here. I am planning to resort to TCP servers when I will be ready with the learning algorithm.