stupiditytax

[stupiditytax]

I’ld like you to meet the disco uk Christmas lotto elf.  He has a very insubstantial body and only three fingers.  He helps to point the way to big winnings and was generated in part by data visualization algorithms and a linguistic neural network.

After many tweaks and long sequences generated the best numbers coming out of the neural network work predicted for Saturday, December 29th.  Given that the prediction came off of a training that was designed for Saturday, December 22nd, I am inclined to savor the imprecision as way to give myself a little family time, and as a way to honor the possibility of some confusion in the predictive powers of the neural network.  In our last attempt using a similar methodology of glitch embracing, we got two numbers right.  That results in a Lucky Dip.  Free numbers that the Lotto itself grabs for us.  Our Lucky Dip numbers for Saturday are 21, 22, 31, 34, 47, 56.  If the neural net numbers are wrong, I am convinced that they were making way for the lotto’s algorithm.  All these scripts are connected to networks of machine learning, and also of inter-computer communication. Would it be so strange if my neural network became aware of the guess that was made in our tiny sphere of the randomly computational galaxy by a behemoth computer in London?  I think not.  So maybe we win a few bucks off of this set? If not you can use the numbers that generated our disco elf up above:

13, 15, 27, 28, 35, 50 It’s a good set.  Only one super recent number a bunch of average age (7) numbers, and one old-timer in the middle.   Plus...  it’s jaunty like a disco elf.  While everyone is celebrating over Christmas and New Years, let’s hold onto this set  through the rest of 2018, and then go back and consult the Oracle again for the January 2nd, 2019 draw.  The bold numbers above are the Stupidity Tax endorsed numbers for the next full week until New Years.

[stupiditytax]

I have been trying a ton of different methods and settings to get the neural network to generate a number that I feel could reasonably be expected to win.  I tried priming the sample with the date in the same format as the text file I have been using, and then I would try without priming the sample.  I kept getting repeat numbers and repeat number combos, and my confidence would be shot. Then I thought about the last draw which generated one matching number across two draws.  Not a big success, but it felt good.  So what if I let the neural net get creative, and see if that’s the matrix sending me a message?  I recently watched Ralph Breaks the Internet with my six-year-old, and decided that, as a fan of Tron,  I could declare RBI as the better successor to the T=original Tron mantle then that Daft Punk soundtrack of a film that came out a few years back.  If you buy that, then why wouldn’t some version of the glitchy princess Vanellope be the hero of my lotto winning dreams. After many passes with and without priming the guess that was generated for the date “Wed 10 Nug 1898″  called out to me.  Nug is short for.....?  December?  no.  Nugember?  Maybe?  and the national lottery was probably not in existence in any form we could imagine today back then.  I really have no good reason for picking the numbers labeled for that date, other than the fact that they make a nice peaky shape.

For Wed 10 Nug 1898  (and also Wed 19 Dec 2018) your winning numbers must be:  9, 12,13, 18, 27, 39   Glitch power FTW!

[stupiditytax]

A long and winding road

I have new numbers and a new data visualization.  This set of numbers came via numerous trainings of my neural network friend using varied parameters, and then a detour into a psilocybin-feuled musical journey to motivate a process of experimental exploration resulting in a set of numbers with a psychedelic power that is unrivaled in nature. Does that mean they will win?  Hard to say.  Nothing about the  UK Lotto seems very natural to me over here in America (not that the American lotteries make me feel any more organic).  I'll detail my process below and let you be the judge.

As I usually do, I ran a number of different parameters on the updated sample of historical lottery numbers via char-rnn.  I found a setting that was pretty strict in it's adherence to the original that generated a number of guesses for the year 2018.  Including a wrong guess for Saturday December 8th… which was in the historical data set.  So I had a primed guess for Wednesday, December 12th   (which I was asking for), and further down the list a weird guess for December 8th (which had just happened).

Now a quick detour. I have recently taken an interest in the concept of micro-dosing psychedelics as there seems to be a surge of information on the subject lately.  In that process, I was introduced to a Spotify playlist compiled for a psychedelic journey.  I figured maybe listening to that playlist minus the actual dosing would maybe average out to a microdose, and ran that playlist in the background as I decided what numbers to pick for December 12th. 

Kelan Thomas - A Playlist for Psylocybin

 Some time around track 11 or 12 during the bouncing uplifting ambient strains of “Peacock Tail” by Boards of Canada, I felt deeply in the zone, and happy about the pick.

Here are the numerological gymnastics that got us to this set that I am about to reveal.

For the December 12th entry the rnn picked the somewhat unlikely sequence of 4, 11, 17, 23, 42, 54.

It seems to always, regardless of settings choose a set that starts with 4 and then 11… sometimes 5 and then 11, but you get the picture.  Thus, I am always skeptical of a 4, 11 set.  But then I saw that at the bottom of the list I had generated the date Saturday, December 8th showed up with the number set 07, 11,  29, 41, 43, 56. This set of numbers was verifiably wrong. So what could I do? I chose to use the December 8th numbers as a sort of Rosetta Stone.

The actual numbers:

06,  23,  36,  38,  43,  48

modified by:

+1, -12, -7, +2, 0, +8

makes the number guessed:

7, 11, 29, 41, 43, 56

So I tried to do a direct correlation.

4, 11, 17, 23, 42, 46

modified by:

+1, -12, -7, +2, 0, +8

makes the number guessed:

5, -1, 10, 25, 42, 54

Well.... as nice a solution as that might have been, a negative one lotto ball is a rare thing indeed, and the dissonance of this result was making my micro trip feel a bit emo and angsty.  I needed a quick fix for that negative number. 

 I ran another python script that listed the current gaps since last picked for each number, and grouped equivalents together based on their age.

The result:

first equivalent gap

(array([4]),)

second equivalent gap

(array([ 16, 20]),)

third equivalent gap

(array([ 0,  6, 36, 38, 43, 48]),)

fourth equivalent gap

(array([ 0,  6, 36, 38, 43, 48]),)

fifth equivalent gap

(array([10, 15, 27, 30, 32]),)

sixth equivalent gap

(array([39, 52]),

I could have decided to choose the number to guess from only the set above, and if a 4 comes up on December 12th I am going to feel sad that I didn't.  I chose to use the equivalent gap only for the negative number.  That left me with either 16 or 20, either of which would move that original 2nd number selection to 3rd since the listing is numerical.  Based on the original third position guess of 17 by the rnn algorithm, I figured 16 was a pretty good guess for third highest number.  I put the new numbers into my graphing script and got the most boring, non-psychedelic, fat lightning bolt graph you could imagine.

The numbers that made up this graph:  5, 10, 16, 25, 42, 54

Happy to see a 50 in there too.

[stupiditytax]

The top image is the actual numbers and the bottom is the guess (wherein only lucky 13 matched).  The symmetry of the actual numbers is mindboggling.

For Saturday the 7th’s UK Lotto... the computer has guessed 7, 12, 27, 33, 45, 49.

45 would be a repeat so you might want to hedge your bets and insert a number in the fifties if you are looking to hit all 6 digits.  At this point I just want to see if I can get to 3 or 4 digits... 5 would be heavenly.  Hell, maybe 45 will be a repeat.  In any case, if the numbers the computer has guessed do appear the shape they will make based on my data visualization algorithm will be as follows.

looks strangely familiar, doesn’t it?  I can’t tell if I am mapping the probabilities of char-rnn or the lotto.  I guess Saturday’s result might tell me?

[stupiditytax]

The image above is the chart mapping the age of each number recently selected in the uk lotto for this past Saturday December 1st.  

This image is the set of numbers that char-rnn neural network chose (with some assistance from me) for tomorrow’s draw Wednesday, December 5th.

The neural network script guessed two numbers right in the Saturday draw, which means we get a “lucky dip!”.  Last time that happened the “lucky dip!” won £25.  The “lucky dip!” is a random number generated by the lotto.  Seems they have a bit of an edge if you ask me.

The above crashing fowl shaped figure strikes me as on track for an extra match or two.  Three maybe four next time, as the AI brain is fed more numbers to compare with.  I am also working toward a different script but having trouble getting certain things to play properly with my Mac OS.  I’ll let you all know if I get that cracking over the holidays. In the meantime play  13, 16, 30, 32, 37, 47 and that goose above us will lay a golden egg.  Right?  BTW... these images are NOT pornographic!

[stupiditytax]

I kinda like the spread of these numbers better than my last set.  Though it keeps starting in the single digits.  Also, every training of the network I did when sampled gave me a 46.  Seems to me 46 is hot right now?

Also.. we’re back to the witches hats... so that has to be good right? The pointy image above represents the following set of numbers. 

5, 20, 25, 31, 43, 46

 If I were a betting man and lived in England I would use them.  I would also maybe use the numbers from my last post.   Oh...  and the 10 that I recommended as a swap out for my last guess?  That showed up in the EuroMillions... but 1 outta five is no winner in my book.  

I still rely on a shady buyer in England to purchase the tickets for this project, though I recently was able to watch the Peter Watkins docu-drama about nuclear war called “The War Game” using a VPN with my location set to London (wonder why it won’t play in the U.S.?).  I have often wondered what would happen if I used that same technique to buy a lotto ticket while comfortably ensconced in the American Southwest?

Anyway.. I’m gonna leave it at the numbers above as the preferred set.  Good luck to any who play!

[stupiditytax]

My machinations with machine learning really failed me on the last draw.  Not a single number was correct.  My fixer in the UK bought that loser of ticket, and then decided to switch things up a bit and buy a Euromillions ticket with all 5 substitute numbers I offered for the last Lotto draw.  As I said when I found out...”With my luck, that’s going to be our big winner.”  So check the results tonight for the 10,15, 30, 32, 43.  Not sure what that’s likely to mean for me, but I’ll certainly be very happy for someone in England. In the meantime, I am continuing to learn and experiment with recursive neural network technology.  The graphic above is not my final guess, but a strong contender, and you should by a ticket if you are inclined, in case I can’t get a final choice ready in time.  The graphic above shows the numbers 3, 11, 17, 18, 41, 46.  The number 46 is an immediate duplicate.  It has been picked almost a third of the time in the last 13 draws.  The net seems obsessed with both low numbers and elevens lately, but as the draws pile up it feels like 11 might be a reasonable guess once again.  I’m still not happy with the convenient distribution of the numbers across the count, but I may just need to swallow my pride and let that stand.  I have another possible result cooking as I write this.  If that turns out more interesting, I will use it.  So stay tuned.

[stupiditytax]

I’ve finally found a training method for char-rnn, that seems to be generating useful or interesting results.  2800 rigorous training routines with improving train_loss numbers, after trying several approaches that included an unprimed generation of 5 million + characters (that did not render an appropriate date, even).  The number 11 seems quite persistent (as were 3,4, and 5)  Once I got 9 as a starter, with so many of the other numbers in the same area as other prior guesses beginning with 4 and 1, I figured we were golden.  Eleven is going to return, it has to.

I don’t know why I am more comfortable with 27 and 42 repeating, especially when 42 would be making it’s 3rd appearance.  I think I am not counting on all 6 being right, but that there will be a repeat (as that seems to be happening a lot lately) and maybe it will be one of these numbers.  and if not we’ll at least get 3 or 4 numbers right, maybe.  I’m not going to advertise this set as a big winner, just that we might get several numbers correct. 

The char-rnn selected set is: 9, 11, 23, 27, 42, 48

alternates for 27 and 42 according to my calculations might be  10, 15, 30, 32, or 43.  If you are inclined to substitute those out or play a number of these combinations, well... anything is possible.  I, of course, will feel an agonizing pain of defeat for not having the confidence to make those substitutions myself if one of those five numbers or more completes a winning combination.  And if BIG LOTTO is reading... only look at the bold numbers, ok?

[stupiditytax]

A google reverse image search picked this image as a match for the data visualization below.  An aerial lift.  That combined with the witchcraft of  my hat interpretation seems encouraging.  No?

[stupiditytax]

I have tweaked the training and sampling methods extensively.  I was discouraged by a constantly recurring series of 4s and 5s and 11s.  Or maybe just nonsense which could maybe be parsed into something.  The image above which kind of continues the witches hat shape, but with some flapping wing brim action, also seems continuous with the previous draw.  Given that the numbers were generated by a neural network, and learning machine, it must be better than a learning human, right?  So these are the numbers I’m sticking with.

3, 21, 26, 33, 45, 54

At the rate we’ve been seeing lately, I am hopeful that we will have 3 numbers in this set... or maybe 4.  But 6 is always a possibility too.  So, umm... Happy American Thanksgiving!  and feel free to send your winnings to me by boat if you feel inclined....  just no blankets, ok?

[stupiditytax]

I missed the boat

After the disappointment of the last draw.  The last draw was a draw wherein I chose a big loser set of numbers, but I chose that set based on its comparative value next to a set that guessed 3 out of 6 numbers.  So I decided to tweak a few parameters and try seeding the results every time.  As a result, I started seeing 5, 11,  ...   and 4, 11, .... every time.... or with some of my more intense settings it got more abstract like “Wed 21 Nov 201832 1 32442049 43 34  35″

what am I supposed to do with that? If I parse it on my own.. maybe I get 24 and 42  which would count as two right.

Another one looked like this:  “Wed 21 Nov 201832 1 334422491444  22″ again.. if I had accepted it as a possibility I might have picked 24 and 42 again.. but 2 right is pretty common... I want at least three.  And last time the 3 right option was spaced and articulated exactly as I had expected.

This is all to say.  I got tangled up in a mess of second-guessing and self-recrimination and managed not to make a guess for tonight’s draw.

last weeks winning numbers made a fascinating shape:

The most recent draw numbers looked like this when they were drawn:

Back to witches hats... hmmmm?

[stupiditytax]

So this is the set of numbers that was actually drawn on Wednesday.  The 4 and the 11 repeated.  This looks nothing like a swan or even a frog for that matter.  Maybe an odd angle on a giant spider walking straight at you.

I am debating whether the coincidental synchronicity between the rejected AI generated seeded numbers (which included the 4 and 11) is an example of even a broken clock being right twice a day, or an instance of the AI starting to understand the patterns through repeated usage.

On several repeated runs after massive trainings, the neural network kept starting with 4 and 11.

Strangely, when using the date I wanted as a seed I got 8 numbers instead of 6.  I could try 26, 34, 42, 51, 53, 56 as a number.  That would look like the image below.

That’s a little more like a swan..... I guess.  A swan dipping it’s head underwater to get a fish or some water plants or ... a frog...   and thus...  a winner?  It has a repeated 53, but the next number has a repeated 11.

I  got some numbers from a different training using a seeded process that seem a little more likely than a repeating 4 AND 11.  They are 5, 11, 17, 23, 42, 46.  Those numbers make an image like the one below.

That seems to be getting back to the days of the witches hat. Or a patriarchy smashing boot/high heel shoe.  But it’s not precisely either one.

I want to doubt the recurrence of an 11.  I also want to doubt the logic of a quick increment to 5 from 4. I had those doubts before and it cost me a Lucky Dip, at least.  While my error last time was because I failed to use the seeded result, I wonder now... will my error be USING a seeded result?  I mean I guess both of these sets are from a seeded result, right?

I’m sticking with the swan... I am a beautiful swan.. a winning swan... 

If it’s good enough for Natalie Portman... I’ll eat some frogs.

Our winning numbers for Saturday 17 November 2018  will be  26, 34, 42, 51, 53, 56 .  These are the neural net’s chosen numbers minus the 4 and 11 that it keeps obsessing over.

[stupiditytax]

Neural net is doing pretty well

The Wednesday draw was (against all odds) a set that started with a 4 again...  but not just that a 4 and an 11  !!! That would earn another lucky dip!  But my human bias said “NO! ... you can’t draw 4 again so soon.  You can’t pick the same set of numbers two times in a row if your algorithm makes any kind of sense.”  Stupid human bias!  

Alright, it seems the seeded result is the only one that matters if we truly trust the neural net.  New numbers will be put into the hopper post haste, and baked into a lovely concoction for Saturday.  HAL is working for us.

[stupiditytax]

So I figured out the graphing code and this is the current figure cut by the numbers from the last draw. Now the 4 is a little off the floor, but the two numbers we got right are on the floor.  Strangely the neural net, when seeded with “Wed 14 Nov 2018″ as it’s starting text, generated the exact same set of numbers as for the last draw. For this to be right, two numbers would have to appear in two draws in a row which I believe is a highly unlikely occurrence.

When the neural net first guessed this set of numbers it looked like this, actually.  The new context for the same numbers is at the top of the page.  In spite of the fact that these numbers do make the letter M which is the first letter of my last name, there is nothing scientific or even highly probable about that.

Looking at probability though generates a few other options.  I had the neural net code generate 10,00 sets of dates and numbers. One time I seeded the result with the date as above, and one time I let it try to use pure probability to figure things out.  

The first number out of the unseeded run was a possibility. It looked like this:

A big dipper of sorts, with a duplicate from the previous draw in the 30s, but nothing about it that made me feel compelled to use it.

With the seeded date, the code stuck to its guns from the previous draw, so I figured a new approach might be preferable.

I did a little sorting of the 20,000 seeded and unseeded results, and I discovered that the date 14 Nov showed up several times. Some of those times it was presented as a Saturday, and others as a Wednesday. Other than the seed text, none of the years matched with the date 14 Nov 2018.  I looked back in the calendar and found that among the generated dates 14 Nov was also on a Wednesday in 2001 and 2007. There was a Wed 14 Nov 2007 in each list, but they did not match, while the date Wed 14 Nov 2001 also occurred in both lists and also matched. Seeded and unseeded neural network runs both guessed Wed 14 Nov 2001 2 25 44 46 48 49.

While 44 is a repeat from last draw, I still feel good about this one. It looks like a crazy swan charging across a pond to defend the honor of its mate, or to gobble a good size frog treat.  If you play these numbers you will get the metaphorical frog treat.  Swans eat frogs, right?

[stupiditytax]

It was Math Night recently at my child’s elementary school. They had a game designed to teach first graders about probability. Imagine that. Green showed up as often as both red and blue combined. Maybe if I use a spinner instead of a neural network my guesses will be better? Though I have it on good authority (the internet) that the last set of numbers got two right which earned a “lucky dip”. So.. if you play my numbers on Wednesday you’ll also have that second set banging around.

Here’s the question.. if the numbers I pick get a lucky dip... and then the randomly drawn numbers in the lucky dip win millions ... can I claim any credit for that millions winner?

[stupiditytax]

Sorry to go dark for a bit.  Between celebrating my birth and fretting over our elections I was duly distracted from number guessing.

One of the things that has consistently frustrated me in the process of trying to guess numbers has been the idea of repetition.  It seems logical that once a number is picked out of a large enough range, that alone would be enough to make it less likely the next time.  Of course, if a number has a very low total count and needs to catch up to the other numbers in the set, that might provide an exception.  Before you remind me that randomness ensures that none of these concerns really matter, I do understand that.  However, the lottery encourages us to look at statistical probability.  To this end, I  try to look at whatever elements I can to statistically justify a number’s occurrence.

I am still using the character based neural network to select numbers, which I think is somewhat flawed because I think it sometimes will choose a number that has been occurring often because it has been occurring frequently in the data set, rather than because it hasn't been used in a while.   

Tonight’s number predicts a duplicate in the number 4.  When I look historically, 4 was last picked in July, and then again about a month earlier.  Its current total number of times picked is on the upper end of the average. So if it is picked twice in a row, it would be surprising but still fairly negligible in terms of it’s effect on things.  Shape-wise when 4 was picked on Wednesday it was in my high probability category.  

All of this is to say the shape above seems to reveal a problem in the drawing algorithm, but I post it anyway because I like it. Tonight’s numbers are: 

4, 11, 17, 23, 41, 44

[stupiditytax]

For tonight

It’s almost 4 pm on November 3rd. It’s my birthday weekend and my family is keeping me away from my processing computer the past few days. So no new numbers for this draw. I still think the Halloween numbers are looking good though. I have always, in my more rational moments believed that all my technical prognostication was a mere hedge against the astronomic odds one faces in playing the lottery. The numbers from Wednesday still offer a hedge.. whether it is more or less than it was on Wednesday, I can’t really say.

I do think, even though I keep picking new sets of numbers lately that the best lottery strategy is probably to pick a few number combinations and stick with them for as long as possible. Let the winning numbers be the moving target.

All that is, perhaps, just defensive justification of my dereliction of duty. But it’s my birthday, so... gimme a break... and play 21, 22, 27, 37, 41, 59 .

And if you win... remember... IT’S MY BIRTHDAY!!!!