Video and other deliverables

You will submit the following to Brightspace (see Schedule for due date). The primary narrative of the project Part 1 will be in the form of Powerpoint slides and a video presentation using those slides. A good target for video length is 10 – 15 minutes.

• file 1 (a zip file):
  • Powerpoint slides accompanying the video
  • a text file containing a video LINK (e.g., a private link on Youtube or a link to a Zoom recording). The slides and video (e.g., a Zoom recording of a team meeting, where you take turns narrating prepared slides that are screen shared) will focus on those aspects of the project that were of your design (and not pre-specified for you), but you still want a self-contained video too, so you won’t omit pre-specified components either — you just won’t make as big a deal about them — the parts of the project that were pre-specified will be connective tissue.
    • Give an intro to the project big picture as you understand it
    • Describe your state quality function, and your justification of it (which can be intuitive, but sources can be cited here too and would be desirable)
      • Summarize the pipeline of scores being used, from state quality, through rewards and schedule success probability, to expected utility (you can reuse the image you were given, and/or create your own)
      • you can include your thoughts, possibly informed by experiments, on certain parameter settings (e.g., why you chose certain values of x_0, k , gamma, etc)
    • Describe how you specialized the modified generic search function
      • state how the frontier is organized (a priority queue by EU or something else, and something other than a strict priority queue)
      • characterize the utility-driven, forward, depth-bounded search strategy that you converged on
        • best first search?
          • coupled with beam search?
        • heuristic depth first search?
        • iterative deepening heuristic depth first search?
    • Give two or more test case(s) that you regard as interesting and informative. Each test case minimally includes
      • an initial state,
        • why the state? Does it illustrate world balance or certain imbalance(s) that might encourage or discourage cooperation?
      • parameter settings
      • selected output schedules (maybe 2-3) for each case, and the EU scores for each state in each schedule
        • observations about these schedules, perhaps the relative frequency of transfers and transforms
      • Interesting observations, both about the example test cases, and perhaps trends across all test cases, such as whether EU is monotonically increasing or decreasing, or neither, over example test cases and other schedules that are not shown in detail.
    • Other issues to include in your presentation might be
      • whether you used hand-crafted macro operators, examples of them, and your assessment of whether they benefited search or not
      • summaries of experimental results, perhaps in graphical or tabular forms, such as
        • how runtime and/or EU of schedules varied with different parameter settings (e.g., depth bound, beam size). If you know/find how to instrument code in Python to determine runtime, please share with the group.
        • a scatter graph on the EU (y axis) of final schedules, with an x axis that is the order in which the schedules were written to file (in an anytime planner, its idea if better schedules are found earlier than poor schedules)
        • I am more concerned that you do meaningful tests, than with how your simulator actually performs on the tests
        • I won’t be more precise than this right now, but a project with no evaluation, typically quantitative, but possibly qualitative, won’t get a maximum score
      • transform templates that you added to the set that was given to you, and citations to any sources
      • interesting observations about connections to the real world that are not addressed earlier (ideally with citations)
    • citations to outside sources that you used (e.g., on transforms and resources, algorithmic variations), to include references to other teams, which will affect your ‘community score’ and theirs (a source of extra credit).
• file 2 (a zip file): well-documented and well formatted Python code, include a READ ME file with directions for running your code and that identifies the file that contains the game_scheduler function, and any other important top-level functions (e.g., one that iterates through parameter settings that you are experimenting with)
• Don’t worry about submitting the following
• file 3 (a zip file): your actual INPUT FILES, including a file for each test case that you illustrated in the video, with appropriate indenting and
• file 4 (a zip file): your OUTPUT FILES, with clear correspondence with input files based on naming conventions, as appropriate
• file 5: A PDF, test_cases_summary, that is a nicely formatted and documented list of test cases modules. This file will be informationally redundant with other files that you are also submitting in Zip file 3 and Zip file 4, but the emphasis here is on ONE easily comprehensible source in which we can look at a human-comprehensible summary of results. This file is the closest thing to a written report that you will submit, so include some connective, explanatory text, and motivate and explain why you designed each test case. Test cases should include the examples that you used in your presentation. Each test case includes
  • initial state
  • various parameter settings, both
    • passed to game_scheduler
      • your_country_name,
      • resources_filename,
      • initial_state_filename,
      • output_schedule_filename,
      • num_output_schedules,
      • depth_bound,
      • frontier_max_size
    • and constant settings within a run
      • x_0,
      • k,
      • gamma
      • failure cost, C
      • any constants involved in your state quality function
  • a list of at least 5 output schedules (and intermediate and final state scores)