hypersoniq's Blog

$12 per week to play PB. Wednesday got a PB match with PP for $16. I play wed. Thru mon. For the week, buying 3 new chances on tuesday or wednesday morning for the next 3 draws. Had a handful of 0+1 hits that generated a bit of padding against the usual losses... $16 was the biggest single hit. Last draw for this week is tonight. I pick a different path each week through the hot/neutral/cold chart on the app. No rhyme or reason. I maintain based on past experience that my red ball picks end up being better than QP, but the white ball picks are just as bad as QP.

Since there is not yet enough history and PA did not update their old game hooks to the database since cash 4 life, I will look to another of the participating states to grab the data. A simple script to print the white balls to csv should suffice. A format like

WB1 - difference - WB2 - difference .... WB 5 will be row 1

difference -          - difference is row 2

Next draw, repeat. 

That will show the changes row to row and column to column. The cash ball will only have the row to row difference.

Cell formatting can color the backgrounds to encode info like O/E.

Nothing written in stone, just searching for things that hold, generalizations based on pure observation.

There may not be a direct pattern in the numbers at face value, but maybe some other information can be used to find some pattern that holds.

Then there are some concepts that I would like to look at such as a "scoring matrix" for combo selection. This would be extremely off the deep end for lottery analysis but nothing has worked so far so why not? It also helps lay the foundation began by follower Markov chains into full blown hidden Markov models. Just now starting to understand the function of HMMs in other applications.

The idea is to look for things, patterns, coincidences... not limited to any one thing. They can be tracked as well... such as if you are looking at the individual situation of odds and evens, check the rolling percentage over several draws. The end goal is to develop a list of things that hold over multiple draws. Things you might otherwise miss if doing bulk statistics over many draws.

You may end up with some odd collections...

Perhaps odd follows odd at a higher percentage than odd follows even... remember, this is per column as well as across the combo.

There could also be some way to re imagine the history, rather than a flat 2D file, it could be a 3D array with the tine series step as the third dimension.

There are no rules here, be creative... regular classical statistics have not been able to "crack the code".

Start with the first 2 draws. Make an exhaustive list for each position how one got to the next, even if it is NOT based on math or statistics. Add the next draw and remove those that failed, keep those that held. Repeat until you are left with a handful of "rules" general enough to have held to that point. Then, and only then can you look for spreadsheet or code automation to push forward.

My goal with every system was always to end up with one single pick as a "best guess". This was, and remains the driving force,  because it creates a minimal budget.

Not that there haven't been coincidental hits, but they are too few and far between to be profitable.

Systems that result in groups of numbers to play per draw are prohibitively expensive, and when they fail to produce a hit in a reasonable time frame, can render a single win as not profitable because of having to cover the expense of the losers... play 1 number on the pick 3 for a week, cost $7... on a hit of $500, profit is $493. Play 10 numbers for a week, cost is $70 and profit drops to $430. Covering just 10% of the numbers, 100 plays per game in a week costs $700, a single win is still a $200 loss. Then there is the logistics of having to print out 700 tickets over a week.

More combinations per game is definitely not the ideal solution. Since some states have 2 or 4 pick 3 draws per day, you can see how this would escalate quickly.

So I have always tried (and mostly failed) to generate that 1 pick. Since I could not even figure out that one pick on a pick 2 with any consistency, I have reduced my $14/week budget down to $12 by only playing one Power Ball ticket (with opt in multiplier and double draw) per draw. I still use my app and alternate between picking from the Follower Markov Chain or from the classifier output, but that is where I ended up... it is boring waiting for that dumb luck coincidence, but it stays on a budget, so if there is a coincidental win, it is the cheapest way I could have obtained it. It is still a $624 per year budget, which means at the usual power play multiplier of 2, it would take 4 hits of 3+1 or 4+0 to cancel the expense (or 2 hits of 3+1 or 4+0 on the second draw where they pay a flat $500)

So far I have not done better than 2+1. The usual outcome being 0+0 to 2+0. I do not have any new alternatives to number selection at the moment.

Not ready to say chasing this hobby was a complete waste of time yet... but it is certainly leaning in that direction.

So, I started on spreadsheets many years ago. I moved to Python (and a few C++) scripts to analyze csv draw histories. They all have the same thing in common... you have to have a solid idea before creating anything. I have taken these game histories across multiple concepts and it takes time to create each one only for the same result... nothing but an occasional coincidental hit.

There are more ideas, but it would be better to have a rapid way to get a feel for how a new idea might pan out without needing to start from clean sheet spreadsheets or scripts for every single idea (and idea variant).

I know I have mentioned it before, but the path may indeed be SQL. It does not even require converting the csv files into tables! So my initial challenge is to pick a database engine to use, which I have, PostgreSQL 18.3 (which is also an excellent choice for computational biology because it handles JSONB) and using a program like LextEdit to directly enter SQL queries. I have to set up the ODBC drivers for this solution, but if it gets too messy, I can default to the PostgreSQL command prompt, like I did when taking DB courses.

I do not know if this will work, because I have not tried it yet, but it seems better than copying an entire spreadsheet and changing core formulas to try an idea variant or a partial rewrite of Python scripts for the same purpose.

Anyone else tried SQL for rapid idea prototyping?

On my classifier, there is a metric called MRA (most recent appearance) which shows how many draws back that a number in that column was drawn. Technically that is the foundational data for gap analysis... number + last appearance.

But what exactly are we looking for? Repeating gap patterns? Average gap size before a number is drawn?

Gap data looks as random as the drawn numbers...

Even in noisy real world bioinformatics data sets there is a structure of biological rules present in the data, even if separated by millions of years of mutations and evolution... lottery data has no guardrails... no underlying truths to build upon.

After seeing it work right in front of me on DNA nucleotides, i know that IF the lottery data could be solved, Markov transition chains would have found it...

That has been the only hit so far after dropping out of everything else. The $12/wk budget is the best part... not winning anything is not so interesting. Not wasting my time on dreaming up some system that might help solve the puzzle is super dull... but when you finally see the truth you can't unsee it. My app now is basically a heavily over engineered QP generator.

Finally got to use a Markov chain in bioinformatics to calculate transition probabilities for Mendellian inheritance, in that domain it is a wonderful tool to have that gets genuine results. I have 39 problems to go in the bio algorithm challenges, they should be introducing hidden Markov models soon. I never quite got past Markov chains with lottery data, so this will be an interesting learning opportunity.

Also, bioinformatics will lead to biostatistics, where I will be getting some good foundational practice using the R language.

Sort of hoping to be inspired to bring something back to this domain, but so far I have seen nothing applicable.

I do stand by my decision to cut back from $14/wk to $12/wk... instantly shaves $104/year off the budget and it is somehow less discouraging to have the chance and miss millions vs missing $500 repeatedly... and it only needs to work once.

Still here, still playing, still playing responsibly.

Taking any number from 2 to n digits, 

1. A sample with 37621, sum the digits (19)

2. Subtract the sum from the original number (37621-19) to get 37602.

3. Sum the digits in the result (18), they will ALWAYS be a multiple of 9! 

Tried this with up to 9 digits and it works every single time.

But what, if anything, can this be used for?

Straight statistical analysis does not work because the history is random, bias, if it exists, is so slight it avoids repeatable detection. 

Most of the systems in the systems thread are variations on the same type of theme... even the "new idea" is 5 years old...

There is not much going on in the math forum.

Outside of a few interesting posts, it is as if there is a lull in the truly novel approaches...

Which leads to the original title... what could possibly be next?

Crazy how the strange tree theorems I learned in college have an actual applied use outside of homework...

So the problem sets up this long winded explanation of unrooted tree theories and wants you to determine the total number of "internal leaves" on the tree... for an unrooted tree, this is returned as n-2... that's it! So basically I downloaded their challenge number, subtracted 2 then resubmitted... this was one of those moments where the CS degree came in handy. I could not imagine having to come at this from the other side, Bio to CS... I am picking up the biology on an as needed basis.

I also learned that the python pow() function accepts a modulo as a third argument... sometimes these biological problems would quickly flood your RAM if you are not thinking of efficiency from the start. Many of the combinatoric solutions want you to use modulo 1,000,000 to keep from freezing your cpu.

"Molecular Biology of the Cell" 7th edition by Alberts et al.

Normally $265 for this hardcover 1,552 text considered the "bible" of cellular biology, Amazon had it at 35% off for some reason... may as well grab it! The research on the field of bioinformatics frequently references this book, as do the complimentary fields of biostatistics and computational biology.

It has been amazing moving from the statistics of useless lottery data to a field where the stats have meaning. I am still sticking with python for the first 103 challenges (33 done now) but the next area, the arsenal, is where the move to R for proper statistical analysis is the right move.

combining python for extract/transform/load and R for statistics and visuals is how they do it in real bioinformatics labs.

Even the lottery app was not a waste as the lessons learned are already making an impact on understanding why these problem sets have biologic importance and how to process and move data through a pipeline from ingestion to final product.

Who knows, once a real exposure to powerful analysis in R moves along, there may be better solutions to this lottery problem that come to light.

Still carrying on with the power ball, one ticket 3x per week, with no good result over the last month... but indeed it only needs to work once!

Book will be delivered tomorrow, $93 cheaper thanks to the "sale"... strange to see a book like this on sale but I am not arguing...

We have even/odd, high/low, so why not try something different.

Set one ("open" numbers) = 1, 2, 3, 5 and 7

Set two ("Closed" numbers) = 0, 4, 6, 8 and 9

Why? Why not?

The open numbers have no enclosed loops, the closed numbers each have 1 enclosed loop, but 8 has 2.

Mixes up vs. High/Low and Even/Odd... could even pair them up like mirrors etc...

Inverted...

1   9

2   8

3   6

5   4

7   0

Have fun...

The last 2 draws, Saturday and Monday, I managed to match exactly one white ball in the double draw... nothing on the main game.

On the 24th attempted challenge at rosalind.info I submitted an answer that did not pass. That was the first fail, and there was much to learn from it. Last night I took another read of the problem and finally figured out what they were expecting, a recursive depth first search of the space provided... then I passed it! I have not dealt with recursion or DFS since my data structures and algorithms classes. Another win for choosing to learn bioinformatics/biostatistics is getting the chance to apply literally everything I had learned in school but did not use since.