Remove enchant #59
Remove enchant #59
Conversation
…due to limited nature of basic word list...
| Initialize the Word Chains game environment | ||
|
|
||
| Args: | ||
| max_turns (int): Maximum number of turns before the game ends in a draw. | ||
| """ | ||
|
|
||
| # Ensure NLTK words are loaded | ||
| self.word_list = list((set(word.lower() for word in words.words()))) |
There was a problem hiding this comment.
technically this behaviour is changed as the dictionary is also built from the other dicts. so not just word stems
| @@ -56,19 +42,21 @@ def reset(self, seed: Optional[int]=None): | |||
| random.seed(seed) | |||
|
|
|||
| # pick a starting word | |||
| starting_word = random.choice(self.word_list) | |||
| starting_word = random.choice(self.dictionary.get_all_words()) | |||
There was a problem hiding this comment.
for prior behaviour set to self.dictionary.nltk_words instead of get_all_words()
| if flag in suffixes: | ||
| # apply suffix | ||
| for rule in suffixes[flag]: | ||
| # continue if flag is not mergeable |
There was a problem hiding this comment.
this code is not very pretty. Fairly certain it works but
There was a problem hiding this comment.
obviously not ideal to add 120,000 LOC but it is just a dictionary file. Need to add sourcing somewhere?
There was a problem hiding this comment.
no effective differences here outside of changing dictionary source
| for i, word1 in enumerate(filtered_words): | ||
| for word2 in filtered_words[i+1:]: | ||
| if self.one_letter_difference(word1, word2): | ||
| # Add edges for words differing by one letter using a more efficient approach |
There was a problem hiding this comment.
more efficient but equivalent version (checked for equivalence)
There was a problem hiding this comment.
efficiency really matters for large word list
| # for i, word in enumerate(self.k_len_words): | ||
| # for other_word in self.k_len_words[i + 1 :]: | ||
| # if sum(a != b for a, b in zip(word, other_word)) == 1: ## check if the words differ by exactly one letter | ||
| # graph.add_edge(word, other_word) |
There was a problem hiding this comment.
remove dead code?
|
|
||
| for start_word in start_words: | ||
| # Use single-source shortest paths from each start word | ||
| lengths = nx.single_source_shortest_path_length(G, start_word) |
There was a problem hiding this comment.
this is more efficient than trying to do shortest path between all nodes... should still give a decent sampling since the words are sampled anyway...
This was really annoying me...
Changes:
Adds dicts for english us and uk (downloaded from hunspell, which is used by e.g. openoffice) (envs.utils.word_lists.py)
Unifies the usage of word lists depending on if NLTK or not, also whether to use pronouns
Removes references to enchant everywhere + dependency
Also does some reworks to WordLadder environment (mainly on algorithm side)
Minor
My parser corrected some things, sorted imports etc.
Also may have to correct some of the typing I used in the word_lists utils file since it uses python 3.11