Jan-Hendrik Ewers Logo

Min-Maxing Your First Wordle Guess

February 6, 2022    wordle python ☕️ buy me a coffee



My pal showed me Wordle for the first time the other day and I instantly wanted to know what the objectively best starting guess was. Some may argue that that’s not in the spirit of the game and that it should be a daily challenge, but those people probably don’t spend 45 minutes on their first game (that’d be super awkward right…?).

TL;DR: oreas, arose, seora, aries, arise, raise, serai, aesir

Firstly we need to figure out some heuristic to measure the goodness of a word. The most intuitive would be to pick a word with the most frequently used letters within the set of 5 letter words. Thus, we can measure a words score using

\[p(\textbf w) = \sum^5_{i=0} s(w_i),\]

which translates to the sum of the frequencies of each individual letter.

Enough maths. Let’s get into the code.

Next, let’s get a word list. I chose this one because it was the first one I found. Simple. We’ll read this into memory, but only the 5 letter words.

words = []
with open('words_alpha.txt') as f:
    for line in f.readlines():
        line = line.strip()
        if len(line) == 5:
            words.append(line)

This results in 15918 possible words. My guess the Wordle backend uses a wordlist similar, if not exact, to this one.

We then need to count all the letters and divide by the total amount of letters to get each letter’s frequency.

big_word = list("".join(words))
letter, count = np.unique(big_word, return_counts=True)
freq = count/len(big_word)

This results in a nice graph with a surprising result.

png Most frequent letters in the set of all 5 letter words

a seems to be the most common letter even though it’s a commonly known fact that e is th most used letter in the english language! This is (probably) because we’re limited to the set of 5 letter words, and no all words.

Another optimisation on the quest for the best starting word, is optimising the word to exclude repeating letters. Having skiing, for example, would be a waste of a letter so let’s employ a nifty hack using set.

print(f"{'e'*9}, {len('e'*9)}, {len(set('e'*9))}")
eeeeeeeee, 9, 1

As you can see above, we can easily filter for words that have multiples if the set object containing that word has a length less than 5.

filtered_words = []
scores = []

for word in words:
    if len(set(word)) == len(word):
        score = np.sum([freq_df.loc[letter] for letter in list(word)])
        scores.append(score)
        filtered_words.append(word)

And just like that we’ve got a tie for the top word. All of these 8 words use the 5 most frequent letters (a, e, s, o, r).

png The top 10 objectively best words to use for your first guess in Wordle

Disclaimer - I am terrible at this game, and these words might be terrible too for various reasons. Use at your own risk.