Wordle Lovers - Play the NYT Daily Game

[Orodruin]

Wordle 1 325 3/6

Spoiler: Google sheet analysis

ALERT (seed):

Ufffff! Things are looking grim ... Luckily, the sheet tells me that DIVER has an expected remaining entropy of 1.62 - corresponding to about 3 words remaining, meaning that at least I should be able to solve in 3 or 4 if things don't go awry.

DIVER:

There it is!

 

[Orodruin]

Other fun things to do with Google Sheets:

Checking the scenario where my standard seed returns all misses. 102 words remaining, but still getting the expected entropy down to 1.62 after the next guess (corresponding to 3.08 words remaining after that guess).

I feel my old strategy of using SCION as the second seed if ALERT does not give much information is somewhat validated by SCION being the third best second guess in this scenario, only slightly out-edged by SPINY and NOISY.

FUZZY and FIZZY would be horrible second guesses ...

 

[jack action]

Wordle 1,325 3/6

[PALER] 39 left
[CUTIS] filter - 1 left
[REVUE]

 

[kuruman]

Assured solution method count update. (Success/Fail): 12/0 - 100.0%/0.0%

 

[OmCheeto]

Wordle 1,325 3/6

 

[gmax137]

Wordle 1,325 5/6







ugh

 

[dwarde]

A little luck involved here

Wordle 1,325 3/6

 

[OmCheeto]

Wordle 1,326 3/6

 

[Orodruin]

Wordle 1 326 3/6

Spoiler: Google sheet analysis

ALERT did not sit well with today's solution. Luckily, SOUTH was determined to be an excellent filter word with an expected remaining entropy of 0.94.

Pretty much spot on expectation, 2 words remaining, BOOTH and TOOTH. Had to pick one!

 

[gmax137]

Wordle 1,326 4/6

 

[kuruman]

Assured solution method count update. (Success/Fail): 13/0 - 100.0%/0.0%

 

[jack action]

Wordle 1,326 3/6

[TRAIL] 12 left
[HONEY] filter - 1 left
[TOOTH]

 

[dwarde]

Wordle 1,326 3/6

 

[Orodruin]

More Google Sheet analytics:

The first table is a number of starting words that I tested for the expected remaining entropy after playing that word given the words that remain after today. Actually surprised to see ALERT do so well in comparison.

The second table are the current (after today) remaining occurrences of all the letters in order of most occurrences to least. E is still by far the most common letter. After EAR it is pretty even among LSOIT so any word with EAR and two from the LSOIT group would be good starting word candidates I suppose. Anybody has any good ones to test?

EDIT: Of course too good to be true. I must have quoted the ALERT result before the computation completed. The correct expected entropy for ALERT is 4.41. Here are some interesting ones selected by using the words with the highest number of remaining occurrences of unique letters in the word:

WOOZY was on the bottom of the list and is a horrible, horrible starting word.

 

[gmax137]

Anybody has any good ones to test?

how about:
CLEAR, then
TOILS

I have been using BLINK as my starter, that stops now!
The frequencies based on your table are 18th, 4th, 7, 9, 20

EDIT: OOps you were asking for just the starter word. Let me think

2nd EDIT:
EARLS, OILER, TREAT (for the double letter)

 

[Orodruin]

how about:
CLEAR, then
TOILS

You mean as a combination of two start words? My sheet doesn't compute that.
I only compute the expected entropy for the start word by itself, then when I insert the result of the start word it computes the expected remaining entropy if choosing each remaining word based on the words remaining. So it takes the result from the seed into account when doing so.

Nevertheless:
CLEAR pops in at 4.58. Not too shabby, but it has already been used as a solution and therefore will most likely never get you that 1.
TOILS is slightly worse as a stand-alone seed, it has an expected remaining entropy of 4.82.

I have been using BLINK as my starter, that stops now!
The frequencies based on your table are 18th, 4th, 7, 9, 20

BLINK: 5.79 in expected remaining entropy. But it can still be a solution since it has not been used yet! And it is still significantly better than WOOZY at 7.14.

Edit: Just to translate entropy to something a bit more tangible: 5.79 in entropy means having almost 3 times as many remaining words as the 4.23 of ARISE.

 

[jack action]

Anybody has any good ones to test?

My first and last 13 words:

4191 arise
4283 saner
4287 arose
4321 least
4344 later
4375 alert
4381 parse
4422 siren
4437 roast
4458 suite
4462 leant
4467 rinse
4506 spare

7894 woozy
7967 pygmy
8001 freer
8114 femme
8135 kappa
8253 slyly
8336 ninny
8428 eerie
9234 mamma
9330 geese
10488 emcee
10520 melee
10612 tepee

I just compute the sum, thus the large number.

SANER and AROSE often change places from day to day.

 

[Orodruin]

I just compute the sum, thus the large number.

SANER and AROSE often change places from day to day.

Google Sheets unfortunately is way too slow to do the full computation for all words and still be viable. I would have to leave it running all day.

I am a bit curious about your word list. According to the one I use there should be 1000 words left after today so the numbers should differ by a factor 1000, which seems approximately true but only approximately ... (That is, assuming you are employing the entropy scoring discussed a couple of months back)

 

[Orodruin]

Wordle 1 327 3/6

Spoiler: Google sheet analysis

Changing to ARISE as starter:

A pretty bad result given ARISE. Only 5 patterns that are worse, including all grays. Still, LEANT seems like a reasonable guess with a pretty good outcome space.

LEANT:

Not too shabby, although more than expected entropy (anything not outright solving it would have). Any of the words will solve the Wordle at most in 4 guesses. Going to have to go with PEDAL.

TADA!

After 1 week of using the spreadsheet, my statistics are:

Pretty decent although statistics are low. Definitely better than my full history, but not clear how much is attributable to the sheet vs other changes made along the way.

 

[jack action]

Google Sheets unfortunately is way too slow to do the full computation for all words and still be viable. I would have to leave it running all day.

I am a bit curious about your word list. According to the one I use there should be 1000 words left after today so the numbers should differ by a factor 1000, which seems approximately true but only approximately ... (That is, assuming you are employing the entropy scoring discussed a couple of months back)

I have a Bash program that takes 30-35 min to compute. I set a Cron job to do it at night so I can use it in the morning.

In the interest of performance - it used to take more than half an hour - I do the calculations for only 237 possible outcomes (even though $3^5 = 243$). Some like should be impossible to get. I might have made a mistake by removing one too many, but I don't think so.

The equation is:

total=$(bc -l <<< "$total + $c * l ($c) / 0.693147181")

Where the function l(x) in the program bc is actually $\ln(x)$.

 

[Orodruin]

In the interest of performance - it used to take more than half an hour - I do the calculations for only 237 possible outcomes (even though $3^5 = 243$). Some like should be impossible to get. I might have made a mistake by removing one too many, but I don't think so.

Indeed, that pattern is of course impossible. However, I compute by word, not by pattern. For each word I check what pattern it would have been sorted into for a given seed and then I make a histogram of how many words fall into each pattern. The additional 6 patterns (why 6, though? with 4 green and one yellow there should be 5 choices for the yellow) are just a simple sum to zero and does not cost much extra computation wise.

 

[OmCheeto]

Wordle 1,327 3/6

 

[jack action]

Wordle 1,327 4/6

[ARISE] 32 left
[GLINT] filter - 5 left: FECAL FELLA HAZEL LEACH PEDAL
[HAZEL] filter + possible answer - 2 left: FECAL PEDAL
[PEDAL]

The suggested 3rd guess was FETCH which would have filtered every possibility. However, I was sure FECAL wasn't the answer, so I went with HAZEL which, at worst left me with FECAL and PEDAL where PEDAL was the obvious choice for me.

why 6, though?

I eliminated all greens as well, which is 1 for every word, so you need to add 1 to each total, but that doesn't change anything relatively speaking.

 

[Orodruin]

[GLINT] filter - 5 left: FECAL FELLA HAZEL LEACH PEDAL

Do you check scores also for words that are not possible solutions? If so, what score did you get from GLINT?

 

[kuruman]

Assured solution method count update. (Success/Fail): 14/0 - 100.0%/0.0%

 

[jack action]

Do you check scores also for words that are not possible solutions? If so, what score did you get from GLINT?

As a seed word, GLINT has a score of 5184.

Otherwise, no, I do not check scores for other guesses. I only do a simple and quick check for the most popular [new] letters for as many letters as possible. I first check the possible answers, then the answer list, and then the word list. It is not perfect. Sometimes, I have a choice of filters and human intervention is necessary. In this case, I only had GLINT, but for my 3rd guess, I had CHAFF and FETCH where clearly it was looking for words with letters CFH. (It most likely first searched for words with DPCFH, then PCFH, and it would have dropped a least popular letter (alphabetically?) until it found a word.)

It is a remnant of my initial program when I did not count scores for my seed word.

 

[kuruman]

Sometimes, I have a choice of filters and human intervention is necessary.

Please help me understand. What is your goal? It sounds like you are perfecting an algorithm that will find the solution without human intervention using as few guesses as possible after eliminating previously used words. Is that right?

 

[jack action]

Please help me understand. What is your goal? It sounds like you are perfecting an algorithm that will find the solution without human intervention using as few guesses as possible after eliminating previously used words. Is that right?

At first, I just wanted to use regexes to determine the best next guess. I thought easy-peasy. Then I included an automatic daily updated list of possible answers. Then I tried to get the best seed word based on that list, which led me to a more math-involved technique.

I just have more fun watching a program I made solve the puzzle than solving it myself. I always prefer watching a machine I made do the work rather than do the work itself. I find it more gratifying as: 1) You need to clearly understand what you are doing (Intellectual labor); 2) You did some work anyway (Manual labor); 3) You are often more efficient. Win-Win-Win.

And then discussing and sharing how I did it with others and finding out how they solved it for themselves is the cherry on top.

 

[Orodruin]

Please help me understand. What is your goal? It sounds like you are perfecting an algorithm that will find the solution without human intervention using as few guesses as possible after eliminating previously used words. Is that right?

That’s - kind of but not really - what I try to do. To select among the possible answers the one that has the lowest expected entropy after the guess, i.e, the lowest logarithm of the number of remaining words.

Technically there will be situations where a word that can be discarded will be a better choice, but that can be kind of handled by hand. It is about making an informed choice.

 

[gmax137]

Wordle 1,327 5/6

SLATE
EARLY
ANGEL
IDEAL
PEDAL

I don't keep track but I feel like my scores are getting worse. I believe as more and more words are used, the game gets harder if you do not look at the list of previous solutions. Looking after the fact, I see that my first three guesses are previous words.