1    Overview

In this second part of the project, we will play the full two-player version of Infexion.  Before you read this specification you should re-read the ‘Rules for the Game of Infexion’document (v1.1 as of this writing). The rules of the game are the same as before, however we have explicitly defined what happens in the (unlikely) scenario where no tokens remain on the board. Note that a log of any changes is kept at the end of the document.

The aims for Project Part B are for you and your project partner to (1) practice applying the game-playing techniques discussed in lectures and tutorials, (2) develop your own strategies for playing Infexion, and (3) conduct your own research into more advanced algorithmic game-playing techniques; all for the purpose of creating the best Infexion–playing program the world has ever seen.

o

We will yet again be using Gradescope as the testing environment when assessing your code. You can (and should) regularly submit to Gradescope as a means to get immediate feedback on how you are progressing.  The autograder will be equipped with simple (not particularly clever) “test”opponent. See the Submission section at the end of this document for details.

The Task

Your task is twofold.  Firstly, you will design and implement an agent program to play the game of Infexion. That is, given information about the evolving state of the game, your program will decide on an action to take on each of its turns (we provide a driver program to coordinate a game of Infexion between two such programs so that you can focus on implementing the game-playing strategy).  Section 2 describes this programming task in detail, including information about how our driver (“referee”) program will communicate with your agent program and how you can run the driver program.

Secondly, you will write a report discussing the strategies your program uses to play the game, the algorithms you have implemented, and other techniques you have used in your work, highlighting the most impressive aspects. Section 3 describes the intended structure of this document.

The rest of this specification covers administrative information about the project. For assessment criteria, see Section 4. For submission and deadline information see Section 5. Please seek our help if you have any questions about this project.

2    The program

You have been given a template Python 3.10 program in the form of a module called agent. Alongside this module is the “driver”program named referee, which is what is used on Gradescope to play two agents against other and enforce the rules of the game. We’ve given this to you so you can test your work locally, but remember to test in the online environment periodically since we’ve provided a simple opponent your agent can verse before the deadline.

o

Before continuing, download1   and extract the template and run it from the root directory via the command python  -m  referee  agent  agent.   This will play the template agent module against itself (naturally this will result in a failed game as it’s not implemented yet!). Details on how to use the referee are specified below.

2.1    The Agent class

When imported, the agent module must define a class named Agent, which contains the following three methods (this has already been set up for you in the template code):

1. def     init   (self,  color:   PlayerColor,  **referee:   dict):  Called once at the be- ginning of a game to initialise your player. Use this method to set up an internal representation of the game state.

The parameter color will be PlayerColor .RED if your program will play as Red, or the string PlayerColor .BLUE if your program will play as Blue. Note that that the PlayerColor enum is imported from the referee .game module – you will see numerous types like this in the template. We discuss the **referee param later on, as this is common to all methods.

2. def  action(self,  **referee:   dict)  ->  Action: Called at the beginning of your agent’s turn. Based on the current state of the game, your program should select and return an action to play. The action must be represented based on the instructions for representing actions in the next section.

3. def  turn(self,  color:   PlayerColor,  action:   Action,  **referee:   dict):  Called at the end of each player’s turn,  after the referee has validated and applied that player’s action to its game state. Use this opportunity to update your own internal representation of the game state.

The parameter player will be the player whose turn just ended, and action will be the action performed by that player. Of course, if it was your agent’s turn that just ended, action will be the same action you returned through the action method. Again, actions will be represented following the instructions for representing actions in the next section. You may assume that the action argument will always be valid since the referee performs validation before this method is called (e.g., your turn method does not need to validate the action against the game rules).

You may optionally use the referee parameter in these methods (strictly speaking this parameter represents keyword arguments as a dictionary, and may be expanded if desired). It contains useful metrics passed from the referee, current as of the start of the method call:

• referee["time_remaining"]: The number of seconds remaining in CPU time for your agent instance. If the referee is not configured with a time limit, this will be equal to None.

• referee["space_remaining"]: The space in MB still available for use by your agent instance, otherwise None if there is no limit or no value is available (only works on Unix systems).

• referee["space_limit"]:  This is a static space limit value available on any system.  It might be handy to have in the __init__ ( . . .)  method if you pre-compute any very large data structures. If no limit is set in the referee, it will equal None.

2.2    Representing actions

We represent all actions using dataclass definitions, as seen in the file referee/game/actions .py. Also note the  “hex representation” dataclass definitions which serve as sub-structures in this context. You can find the definitions of these in the file referee/game/hex.py. Importantly, the same axial coordinate system that was used in Part A is also used here – we simply aren’t using raw tuples to represent actions anymore.

• To construct a SPAWN action use:

SpawnAction(HexPos(r ,  q))

The HexPos constructor arguments r , q denote the coordinate of the token being spawned on the game board. Please carefully revisit the game rules if you haven’t already, as Part A did not familiarise you with SPAWN actions.

• To construct a SPREAD action use:

SpreadAction(HexPos(r ,  q),  HexDir((dr ,  dq)))

The arguments r , q denote the coordinate of the SPREAD action origin, and the tuple (dr , dq) represents the direction using one of the six hex offsets defined in part A. Alternatively, you can use some more “readable” HexDir enum entries:

SpreadAction(HexPos(r ,  q),  HexDir .<dir>)

...where dir may be one of DownRight, Down, DownLeft, UpLeft, Up, or UpRight. The naming of these assumes the default board orientation in the game specification. You can see the exact mapping of these to the underlying hex offsets in the referee/game/hex.py file.

2.3    Running your program

To play a game of Infexion with your agent module, we provide a driver program—a Python module called referee which sits alongside it in the template. You don’t need to fully understand how the referee program works under the hood (suffice to say parts of it are quite complex).

However, it’s important that you are aware of the high-level process it uses to orchestrate a game between two agent classes, summarised as follows:

1. Set up a Infexion game and create a sub-process for each player’s agent program.  Within each sub-process, instantiate the specified agent classes for each of Red and Blue, as per the command line arguments (this calls their  .   init   () methods).  Set the active player to Red, since they always begin the game as per the rules.

2. Repeat the following until the game ends:

(a) Ask the active player for their next action by calling their agent object’s .action( . . .)

method.

(b) Validate the action and apply it to the game state if is allowed, otherwise, end the game

with an error message. Display the resulting game state to the user.

(c) Notify both agent objects of the action by calling their .turn( . . .) methods.

(d) Switch the active player to facilitate turn-taking.

3. After detecting one of the ending conditions, display the final result of the game to the user.

To play a game, the referee module (the directory referee/) and the module(s) with your Agent class(es) should be within your current working directory. You can then invoke the referee, passing the respective modules as follows:

python  -m  referee  <red module>  <blue module>

...where <red module> and <blue module> are the names of the modules containing the classes to be used for Red and Blue, respectively.  The referee comes with many additional options to assist with visualising and testing your work. To read about them, run‘python  -m  referee  --help’.

2.4    Program constraints

The following resource limits will be strictly enforced on your program during testing.  This is to prevent your agent from gaining an unfair advantage just by using more memory and/or com- putation time. These limits apply to each player agent program for an entire game. In particular, they do not apply to each turn separately. For help measuring or limiting your program’s resource usage, see the referee’s additional options (--help).

• A maximum computation time limit of 180 seconds per player, per game.

• A maximum memory usage of 250MB  per  player,  per  game  (not including imported libraries).

You must not attempt to circumvent these constraints. For example, do not use multiple threads or attempt to communicate with other programs or the internet to access additional resources.

2.5    Allowed libraries

Your program should use only standard Python libraries, plus the optional third-party library NumPy. With acknowledgement, you may also include code from the AIMA textbook’s Python library, where it is compatible with Python 3.10 and the above limited dependencies. Beyond these, your program should not require any other libraries in order to play a game .

o

If in doubt, you can always check your work on Gradescope for extra reassurance that the libraries you are using will be compatible with our assessment tools.

However, while you develop your agent program, you are free to use other tools and/or programming languages. This is all allowed only if your Agent class does not require these tools to be available when it plays a game.

For example, let’s say you want to use machine learning techniques to improve your program. You could use third-party Python libraries such as scikit-learn/TensorFlow/PyTorch to build and train a model. You could then export the learned parameters of your model. Finally, you would have to (re)implement the prediction component of the model yourself, using only Python/NumPy/SciPy. Note that this final step is typically simpler than implementing the training algorithm, but may still be a significant task.

3    The report

Finally, you must discuss the strategic and algorithmic aspects of your game-playing program and the techniques you have applied in a separate file called report .pdf.

This report is your opportunity to highlight your application of techniques discussed in class and beyond, and to demonstrate the most impressive aspects of your project work.

3.1    Report structure

You may choose any high-level structure of your report. Aim to present your work in a logical way, using sections with clear titles separating different topics of discussion.

Below are some suggestions for topics you might like to include in your report.  Note that not all of these topics or questions will be applicable to your project, depending on your approach – that’s completely normal. You should focus on the topics which make sense for you and your work. Also, if you have other topics to discuss beyond those listed here, feel free to include them.

• Describe your approach: How does your game-playing program select actions throughout the game?

Example questions: What search algorithm have you chosen, and why? Have you made any modifications to an existing algorithm?  What are the features of your evaluation function, and what are their strategic motivations?  If you have applied machine learning, how does this fit into your overall approach? What learning methodology have you followed, and why? (Note that it is not essential to use machine learning to design a strong player)

• Performance evaluation: How effective is your game-playing program?

Example questions: How have you judged your program’s performance? Have you compared multiple programs based on different approaches, and, if so, how have you selected which is the most effective?

•  Other aspects: Are there any other important creative or technical aspects of your work?

Examples:  algorithmic optimisations, specialised data structures, any other significant effi- ciency optimisations, alternative or enhanced algorithms beyond those discussed in class, or any other significant ideas you have incorporated from your independent research.

•  Supporting  work:  Have you completed any other work to assist you in the process of developing your game-playing program?

Examples: developing additional programs or tools to help you understand the game or your program’s behaviour, or scripts or modifications to the provided driver program to help you more thoroughly compare different versions of your program or strategy.

You should focus on making your writing succinct and clear, as the overall quality of the report matters. The appropriate length for your report will depend on the extent of your work, and how novel it is, so aiming for succinct writing is more appropriate than aiming for a specific word or page count, though there is a hard maximum as described below.

Note that there’s probably no need to copy chunks of code into your report, except if there is something particularly novel about how you have coded something (i.e., unique to your work). Moreover, there’s no need to re-explain ideas we have discussed in class.  If you have applied a technique or idea that you think we may not be familiar with, then it would be appropriate to write a brief summary of the idea and provide a reference through which we can obtain more information.

3.2    Report constraints

While the structure and contents of your report are flexible, your report must satisfy the following constraints:

• Your report must not be longer than 6 pages (excluding references, if any).

• Your report can be written using any means but must be submitted as a PDF document.

4    Assessment

Your team’s Project Part B submission will be assessed out of 22 marks, and contribute 22% to your final score for the subject. Of these 22 marks:

•  11 marks will be allocated to the performance of your final agent (you can only submit one, so pick your best if you developed a few agents).

Marks are awarded based on the results of testing your agent against a suite of hidden‘bench- mark’ opponents of increasing difficulty, as described below.  In each case, the mark will be based on the number of games won by your agent. Multiple test games will be played against each opponent with your agent playing as Red and Blue in equal proportion.

5 marks available:  Opponents who choose randomly from their set of allowed actions each turn, or use some form of weighted random distribution to pick moves.  The currently visible tests provide an example of such an agent.

3 marks available:  ‘Greedy’opponents who select the most immediately promising action available each turn, without considering your agent’s responses (for various definitions of‘most promising’).

3 marks available:  Opponents using any of the adversarial search techniques discussed in class to look an increasing number of turns ahead.

Note that a significant portion of marks are awarded based on performance against a “ran- dom” opponent. Hence, your first goal should be to simply create an agent that consistently completes games without syntax/import/runtime errors, without invalid actions, and without violating the time or space constraints.

The tests will run with Python 3.10 on Gradescope.  Programs that do not run in this environment will be considered incorrect and receive no marks for performance. Like in Part A of the project, you should submit to Gradescope early and often – you can already test your work against an agent which is live now.  While the agent is not very smart, it reliably plays valid actions!

•  11 marks will be allocated to the successful application of techniques demonstrated in your work.

We will review your report (and in some cases your code) to assess your application of adversar- ial game-playing techniques, including your game-playing strategy, your choice of adversarial search algorithm, and your evaluation function.  For top marks, we will also assess your ex- ploration of topics beyond just techniques discussed in class. Note that your report will be the primary means for us to assess this component of the project, so please use it as an opportunity to highlight your successful application of techniques.  For more detail, see the following rubric:

0–5 marks: Work that does not demonstrate a successful application of important techniques discussed in class for playing adversarial games.   For example, an agent just makes random moves would likely get 0 marks. 

6–7 marks: Work that demonstrates a successful application of the important techniques discussed in class for playing adversarial games, possibly with some theoretical, strategic, or algorithmic enhancements to these techniques.

8–9 marks: Work that demonstrates a successful application of the important techniques discussed in class for playing adversarial games, along with many theoretical, strategic, or algorithmic enhancements to these techniques, possibly including some significant enhancements based on independent research into algorithmic game-playing or original strategic insights into the game.

10– 11 marks: Work that demonstrates a highly successful application of important tech- niques discussed in class for playing adversarial games,  along with  many  significant theoretical, strategic, or algorithmic enhancements to those techniques, based on in- dependent research into algorithmic game-playing or original strategic insights into the game, leading to excellent player agent performance.

As per this marking scheme, it is possible to secure a satisfactory mark by successfully applying the techniques discussed in class. Beyond this, the project is open-ended. Every year, we are impressed by what students come up with.  However, a word of guidance:  We recommend starting with a simple approach before attempting more ambitious techniques, in case these techniques don’t work out in the end.

A

In the rare case where there is a disagreement between students about their contri- butions to the project which they cannot resolve within the group, we may assign a different mark to each team member. Please notify us early on if you are having issues with your project partner, and keep records of your attempts to communicate.

4.1    Code style/project organisation

While marks are not dedicated to code style and project organisation, you should write readable code in case the marker of your project needs to cross-check discussion in your report with your im- plementation. In particular, avoid including code that is unused. Report marks may be indirectly lost if it’s difficult to ascertain what’s going on in your implementation as a result of such issues.

4.2    Academic integrity

Unfortunately, we regularly detect and investigate potential academic misconduct and sometimes this leads to formal disciplinary action from the university. Below are some guidelines on academic integrity for this project.  Please refer to the university’s academic integrity website 2  or ask the teaching team, if you need further clarification.

1. You are encouraged to discuss ideas with your fellow students, but it is not acceptable to share code between teams, nor to use code written by anyone else.  Do not show your code to another team or ask to see another team’s code.

2. You are encouraged to use code-sharing/collaboration services, such as GitHub, within your team. However, you must ensure that your code is never visible to students outside your team.  Set your online repository to ‘private’mode, so that only your team members can access it.

3. You are encouraged to study additional resources to improve your Python skills. However, any code adapted or included from an external source must be clearly acknowledged . If you use code from a website, you should include a link to the source alongside the code. When you submit your assignment, you are claiming that the work is your own, except where explicitly acknowledged.

4. If external or adapted code represents a significant component of your program, you should also acknowledge it in your report.  Note that for the purposes of assessing your successful application of techniques, using substantial amounts of externally sourced code will count for less than an original implementation.  However, it’s still better to properly acknowledge all external code than to submit it as your own in breach of the university’s policy.

5. If you use LLM tools such as ChatGPT, these must be attributed like any other external source – you should state exactly how you’ve used them in your report (under “References”). Technology to detect use of such tools is constantly evolving, and we will endeavour to use what is available come marking (or even retrospectively) to detect dishonest use of it. We do, however, believe such tools can be useful when used in the context of proper understanding of a subject area – in short, use them responsibly, ethically, and be aware of their limitations! 

5    Submission

o

The  deadline  is  11:00PM  on  Wednesday  the  10th    May,  Melbourne  time (AEST). You may submit multiple times – only the latest submission will be marked.

The procedure for submission via Gradescope is almost identical to that of previous submissions in this subject. Once again, please include the exact same team .py file in your team’s submission.

Note that only one team member needs to submit the final work.  Once submitted, they must then  add  their  partner  to  the  submission  within  the  Gradescope  interface .   If team members submit individually we will randomly pick one of the submissions to mark and link team members based on information in the team .py file. Projects won’t be remarked if the “wrong”one

was picked in such a scenario (i.e., if one was an old submission).

Here’s how the directory structure for your submission should look:

Please don’t submit the referee with your final submission as it will only serve to clutter it. More generally, marks may be lost if your submission is notably disorganised, so please take care.

You may submit multiple times on Gradescope, and you are in fact strongly encouraged to do this early and often in order to test your work in the assessment environment. If you do make multiple submissions, we will mark the latest submission made.

A

Late submissions will incur a penalty of two marks per day (out of the 22 total marks allocated for this part of the project).