CS 391 Home Page

CS 391 - Selected Topics: Game Artificial Intelligence
Fall 2025

Course Information

Course Information Page

Class Details

Class 1 ():
- Preparation: Creation of BoardGameArena.com account (no subscription necessary)
- Course Overview
- Introduction to the game of Saint Petersburg (SP)
  - Rules (1st edition)
  - Means of playing:
    - Print-and-Play: I’m willing to print card sheets for students enrolled in this course, and I have the tools and can teach the skills necessary to create your own playable set. Enrolled students can email me and I’ll be glad to help create a physical copy of the game in office hours. The only cost to you if you want an easy shuffling experience would be to purchase clear plastic card sleeves, specifically at least 120 sleeves of this size. Other options are listed here. (All money and points can be easily tracked with a free app such as Score Counter for Android or Score Anything for iOS. Here are other recommendations for score-keeping apps.
    - Guenther Rosenbaum Windows application: There is a free, fan-made, single-player vs. AI-only Saint Petersburg application (see email for link) by Guenther Rosenbaum that can be played using cross-platform software called Wine. Professor Presser has generously taken the time to pre-install it as an easy-to-launch application on all lab machines (“Saint Petersburg”), but you can also install this on your own personal computer.
    - Board Game Arena: Saint Petersburg is still a popular-enough game to have regular players online at Board Game Arena 21 years after its release. To initiate a game, one must have a subscription, but anyone can play for free who is willing to wait for an open table (or an invitation from a paid subscriber). There were 505 games in progress at time of writing.
  - Strategy discussion can be found at the BGG forum.
Class 2 ():
- Preparation: play of SP, observing significant features for decisions
- Java Software Overview, Feature Engineering, Interaction Terms
- Raw Features vs Engineered Features
  - Demonstrations:
    - Circular Classification - Add X1^2 and X2^2 derived features
    - XOR Classification - Add X1X2 derived feature (interaction term)
  - Video: What is Feature Engineering? (Microsoft Learn)
  - Reading:
    - GeeksforGeeks: What Is Feature Engineering?
    - Derived (Engineered) Feature Example
  - Further reading (optional): Wikipedia: Feature Engineering
- Interaction Terms
  - Video: Econometrics Interaction Terms
  - Reading: Analytics Vidhya: A Guide to Understanding Interaction Terms
  - Further reading (optional): Wikipedia: Interaction (statistics)
Class 3 ():
- Preparation: Moodle work log, complete new player with additional features
- Logistic and Linear Regression, Data Acquisition
- Linear Regression
  - Video: Linear Regression in 3 Minutes (3-Minute Data Science)
  - Article: Google ML Crash Course: Linear Regression Section
  - Further reading (optional):
    - Wikipedia: Linear Regression
    - LibreTexts / OpenIntro: Introduction to Linear Regression
- Logistic Regression
  - Video: Logistic Regression in 3 Minutes (3-Minute Data Science)
  - Article: Google ML Crash Course: Logistic Regression
  - Further reading (optional):
    - Wikipedia: Logistic Regression
    - LibreTexts / OpenIntro: Introduction to Logistic Regression
Class 4 ():
- Preparation: Complete new player that seeks to maximize score difference using linear regression on score difference
- Common Reasons for Linear Regression Failure, Reinforcement Learning, Batch Learning
- Common Reasons for Linear Regression Failure
  - Model simplicity, e.g. a line trying to predict a parabola. Often, a linear model cannot capture the complexity of what we seek to model.
  - Feature problems, irrelevant features given significance from limited data, missing relevant features (raw or engineered)
  - Violating assumptions of linear regression, e.g. behavior of learned function well outside of our limited random play data distribution.
  - Video: 3.5 Pitfalls in regression
- Reinforcement Learning
  - Video: Reinforcement Learning: Crash Course AI #9
  - Reading: Sutton & Barto: Reinforcement Learning: An Introduction (2nd ed.) Sections 1.11.6, 3.1
  - Further reading (optional):
    - Google ML Glossary: Reinforcement Learning
    - Wikipedia: Reinforcement Learning
- Batch Learning
  - Reading: Batch (Offline) learning vs Online learning in Artificial Intelligence
Class 5 ():
- Preparation: Moodle work log, experiment with (1) new objective function (e.g. logistic regression win probability breaking ties with score difference), and/or (2) generate better play data for learning from self-play of flat MC player. Note: Feel free to speed collection of play data by having the flat MC player collect fewer samples per action.
- Decision Trees
  - Video: Decision and Classification Trees, Clearly Explained!!!
  - Reading:
    - Decision Tree Algorithm overview explained (TowardsMachineLearning.org)
    - Wikipedia: Decision Tree Learning
  - Further reading (optional): Codecademy: Decision Trees for Classification and Regression
Class 6 ():
- Preparation:
- Bias-Variance Tradeoff, Training and Test Sets, Decision Tree Regularization, Decision Tree Ensembles
- Readings:
- Video: StatQuest: Random Forests Part 1 - Building, Using, and Evaluating
Class 7 ():
- Preparation: Moodle work log, complete Competition 1 entry
- Competition 1 entry due via email
Class 8 ():
- Preparation: Presentation of Competition 1 entry
- Presentation and discussion of entries (1/2)
Class 9 ():
- Preparation: Moodle work log
- Presentation and discussion of entries (2/2)
- Feature Engineering: expected rounds remaining, expected score difference
Class 10 ():
- Preparation: (none)
- Neural Networks, Tensorflow Playground, Jupyter notebook demonstrations, relationship to Logistic regression, linear inseparability of XOR, multilayer feedforward networks, backpropagation
Class 11 ():
- Preparation: Moodle work log
- Feature Engineering: needed inputs to predict expected rounds remaining
- Work session
Class 12 ():
- Preparation: (none)
- Work session
Class 13 ():
- Preparation: Moodle work log
- Work session
Class 14 ():
- Preparation: Moodle work log, complete Competition 2 entry
- Competition 2 entry due via email
- Monte Carlo Tree Search
Class 15 ():
- Preparation: Moodle work log,
Class 16 ():
- Preparation:
Class 17 ():
- Preparation: Moodle work log,
Class 18 ():
- Preparation:
Class 19 ():
- Preparation: Moodle work log,
Class 20 ():
- Preparation:
Class 21 ():
- Preparation: Moodle work log, complete Competition 3 entry
- Competition 3 entry due via email
Class 22 ():
- Preparation:
Class 23 ():
- Preparation: Moodle work log,
Class 24 ():
- Preparation:
Class 25 ():
- Preparation: Moodle work log,
Class 26 ():
- Preparation:
Class 27 ():
- Preparation: Moodle work log,
Class 28 ():
- Preparation:
Final: Final report and Moodle work log due Sunday, December 7th, 4:30PM.

Online Resources

Smile Java Machine Learning Library (Documentation)