GHOST

Player Interface

AI:

• Brute Force AI

---

• Smarter Brute Force AI

---

• Bayesian Inference AI

---

---

• GHOST

---

• A simple word game where players are tasked with choosing a letter such that it forms a valid word fragment from a given dictionary, but also avoids forming a full word [lose case]. Every time a player loses a round, they accrue a letter, and are knocked out of the game if they reach the code word «not dissimilar to the game HORSE»

[Read More][ghost] [ghost]: https://en.m.wikipedia.org/wiki/Ghost_(game)

---

• Player Interface

---

• A class that handles player interaction with the game.
• Uses modification of $stdin and $stdout to allow expansion of the class to accommodate user input from outside the command line

---

• AI:

---

• So far, I have implemented 5 different Ai classes:

Ai::BruteForce | Brute Force AI

This AI executes a dumb brute force algorithm to eventually find a valid play with no regard to result.

• This algorithm has no concern for presenting unique responses, therefore could have a worse case time complexity of O(∞)

[Read More][brute force] [brute force]: https://en.m.wikipedia.org/wiki/Brute-force_search

Ai::Memorizer | Smarter Brute Force AI

Ai::BetterMemorizer | Smarter and Lighter Brute Force AI

This AI extends the previous, but maintains a dictionary file of fragments that it knows are valid with no regard to the result

• This algorithm is only faster than it's predecessor if it is encountering a fragment it has seen before.
• As it ages through it's life cycle, this algorithm grows faster than Ai::BruteForce, but just as likely to lose.

[Read More][faster brute force] [faster brute force]: https://en.m.wikipedia.org/wiki/Brute-force_search#Speeding_up_brute-force_searches

Ai::Bayesian | Bayesian Inference AI

Ai::BayesianSpeedy

Ai::BayesianWinner

These AIs keep a record of it's outputs that have resulted in valid word fragments, and the number of times that playing a letter resulted in a loss. It then takes those values and calculates an optimized response.

• This algorithm will develop a disposition to a certain response
• Due to this, over time, these AIs outperform the preceding AIs in either speed or accuracy
• the downside is the AIs develop predictable behavior that is harder for Ai::BruteForce to defeat, but easier for a human player to defeat.

[Read More][bayesian] [bayesian]: https://en.m.wikipedia.org/wiki/Bayesian_inference

---

Todo

• Extend existing AI classes to make smarter AIs
• Refactor the Player Interface into an agnostic agent
• Split this project into smaller focused projects