An AI that tries to solve words given by a user, hangman-style.
Note: This repository only explains the guessing mechanism, not the mechanism of the hangman game itself. This is also my first repository so please feel free to suggest possible improvements.
Download both "Hangman AI.py" and "dictionary.json", and make sure they are in the same directory. Then simply execute "Hangman AI.py".
The first thing the AI does is get a pool of words with the length of the given word.
It then identifies the most popular letter in said word pool then guesses that letter. Note that it will only count letters appearing multiple times in a single word once (e.g The "p" in happy only counts as 1).
Each time the AI guesses, it stores the letter in a "guessedLetters" list and makes sure it doesn't guess that letter again!
After guessing, it narrows down the wordpool using two pruning methods: False Pruning and Index Pruning.
(I only came up with the names for myself)
- Takes in a letter then removes all words that do not have a certain letter.
- Used when guess is wrong.
- Takes in which letters have already been revealed and their indices. It then removes all words that do not have those letters in their respective indices.
- Used when guess is right.
This process then loops back to the beginning where it finds the most popular letter in the updated word pool.
The words used for this AI is from this word list by dwyl.
All the words are stored in a json file with keys ranging from "1" to "20". The value of each key is a list containing words with the length equal to the key (e.g Key "4" contains all words with 4 letters etc.)
This is done this way so that when the user enters a word, the AI can just get the length of that word, use it as a key, then grab the word pool corresponding to that key. This speeds up the initial process of getting all possible words from a giant word pool.
The mechanism depends on the letter popularities of a word pool. It has 5 steps:
- Get the popularity of each letter.
- Inverse each popularity score.
- Calculate each word's "badness" score by summing the inversed popularity score of each of its letters (letter duplicates won't count multiple times).
- Get the three words with the highest badness scores.
- Display them!
This is my first time posting to Github so I hope I didn't do anything too badly. Just as a sidenote, I added time.sleeps to my program so that it is easier to see what the AI is doing.
The sleeps get faster once the word pool gets narrowed down to 10 or less words. For the record, I don't think this is necessarily a good practice. I just did it this way to make the program perform nicer :)
Thank you for taking the time to read!