Wordle Solver

gistfile1.txt

import os
from urllib import request 
import numpy as np
import pandas as pd

# grab a large corpus of words, sorted by usage counts (Peter Norvig)
if not os.path.isfile('count_1w.txt'):
    request.urlretrieve("https://norvig.com/ngrams/count_1w.txt",
                        "count_1w.txt")

# all valid Wordle words are 5 letters, not proper nouns (no capitals) - take top 5k
with open('count_1w.txt') as file:
    words = list(w for w, _ in (line.strip().split('\t') for line in file) if len(w) == 5)[:5_000]

# or use the dict
# words = [w.strip() for w in open("/usr/share/dict/words", "r") if len(w) == 6 and w.lower() == w]

# stick in pandas series for speed
words = pd.DataFrame(index=words)
words['word'] = words.index  # hamlet

print(len(words), "words loaded")


def wordle_pattern(guess, target):
    out = ['.'] * 5
    counts = {}
    # identify matches
    for i, (a, b) in enumerate(zip(guess, target)):
        if a == b:
            out[i] = 'g'
        else:
            counts[b] = counts.get(b, 0) + 1
    # identify present but wrong place
    for i, (a, b) in enumerate(zip(guess, target)):
        if a != b:
            if counts.get(a, 0) > 0:
                out[i] = 'y'
                counts[a] -= 1
    return "".join(out)


niters = 100

words['score'] = np.log(len(words)) # prior of being uninformative

# by hand filtering as proof of concept - target from 2022-01-06
#
# words['pattern'] = [wordle_pattern('aisle', target) for target in words.word]
# words = words[words.pattern == 'y..y.']
# 
# words['pattern'] = [wordle_pattern('ultra', target) for target in words.word]
# words = words[words.pattern == '.y..y']
# 
# words['pattern'] = [wordle_pattern('waldo', target) for target in words.word]
# words = words[words.pattern == '.gy..']
# 
# words['pattern'] = [wordle_pattern('naval', target) for target in words.word]
# words = words[words.pattern == 'yg.gg']
#
# print(len(words), "words filtered")

for iterx in range(niters):
    # choose a random word for target
    target = words.word.sample(1).iloc[0]

    # generate the pattern for each word as a guess
    words['pattern'] = [wordle_pattern(guess, target) for guess in words.word]

    # count how large each pattern group is
    pat_count = words.pattern.value_counts()

    # transform to entropy (log pattern group size)
    log_pat_count = np.log(pat_count)

    # add to score
    words['score'] += words.pattern.map(log_pat_count)

print(words.sort_values("score").head(20))