I have implemented a learning agent and my agent only wins if more than 250 games are done; if it's less than that, the satisfactory agent wins. If I run 10000 games in the tournament, my agents will lose the first 250 but win the rest of the 9750 ga...

Yes. All games in simulation will adhere to the rules of the game as specified, including that it says "5-10 players". Team sizes and required betrayals have not even been specified for outside this range.

ANONYMOUS wrote I am not intending to provide any fixed rules about what combinations will be tested. Note that assessing your own agent's performance is explicitly part of this project. Given that your agent is not receiving any different treatment ...

ANONYMOUS wrote Definitely looks like an interesting article, but since it is focused on deep learning, I want to make it clear that that is not a part of the content of this unit. This article is probably extremely helpful to anyone doing CITS5017 D...

When our agents are being assessed, for single game will the max amount of players always be 10 as specified by the game rules?

I was wondering how the testing would be done with our agents. Currently, I am testing my agent against a pool of agents consisting of my agent, the satisfactory, basic and the random agent in the agent pool. I ask this because I've noticed that my ...

Hi, As this week's lecture is on reinforcement learning, I have a great blog to share from Andrej Karpathy (cofounder of OpenAI) on reinforcement learning https karpathy.github.io 2016 05 31 rl Sharing because this may be useful to Gozz and other s...

ANONYMOUS wrote We do intend to run more samples when assessing your submissions, but 1000 games is still a lot, so if your agent is not consistently winning with 1000 games, it is probably a close enough thing that it's not "consistent". So submittin...

ANONYMOUS wrote No, you should not assume that your agent will survive between games. The testing code provided is not exactly the same as how your code will be assessed, and game outcome is provided so your agents as you are developing them can kn...

Hi, I was wondering what is meant by "consistently outperforms" when marking? For example, my current (not final) agent beats the satisfactory agent consistently every time over the long term when I run tournaments of size 10,000 or 100,000, but in th...

I've noticed that the agent seems to persist throughout the tournament, since it is only created once at the start. This means that it could remember things between games in the tournament. When the projects are marked, are they tested using the same ...

ANONYMOUS wrote Short answer This implementation detail is irrelevant to your task, and you should just ignore it. Long answer I did not write this code, but based on what it is doing this is how it is picking the agents to use in each game. A game c...

in the run tournament.py code we have generate agent objects for use in games number of duplicates math.ceil(10 len(agent classes)) for agent cls in agent classes for i in range(number of duplicates) agent pool.append(create agent(age...

ANONYMOUS wrote Absolutely. Some students seem to have hyper-focused on the warning in the project and misinterpreted it. It says WARNING The reference agents BasicAgent and SatisfactoryAgent have been provided to enable you to assess your code, but ...

Hey, if we wanted to try to make some kind of learning agent are we allowed to use the run game run tournament code. I intuitively want to think we can, but just wanted to make sure.

ANONYMOUS wrote Your agent will be tested (as in the tournament code) against various combinations of agents. It is not valid to assume that any particular agent (or any agent at all) will play randomly. Furthermore, if you were to do so, this would s...

ANONYMOUS wrote You are explicitly required to subclass the Agent class from agent.py . My previous statements have advised students not to analyse the benchmark agents provided (BasicAgent, SatisfactoryAgent), as you should be solving the problem yo...

Hi, For future reference. Please make a new thread for a new topic such as this, rather than extending an old one. Your agent will play games as both spy and loyalist. Your agent's win rate is what is being assessed. It would not make any sense for an ...

Hi, I would like to clarify about your rubric stating "Consistently outperforms hidden reference agent", and I have assessed the code of "random agent.py", where there is a "def is spy(self)", do you expect "spies to outperforms the resistance, highe...