hypersoniq's Blog

Here is what my follower program outputs...

Pick 3 Mid Day Followers
last drawn A =  0
1.0    82
2.0    82
6.0    81
5.0    78
4.0    77
8.0    75
3.0    75
0.0    74
9.0    72
7.0    64
Name: A, dtype: int64

[1.0, 6.0, 8.0, 6.0, 2.0, 7.0, 4.0, 4.0]
The number of followers of 0 is 761
The number of followers derived from the observations is 8
Mode of most recent followers set is 6.0

last drawn B =  8
2.0    86
1.0    82
5.0    79
8.0    77
0.0    76
6.0    73
7.0    73
3.0    72
9.0    65
4.0    60
Name: B, dtype: int64

[8.0, 4.0, 6.0, 5.0, 3.0, 7.0, 5.0]
The number of followers of 8 is 744
The number of followers derived from the observations is 7
Mode of most recent followers set is 5.0

last drawn C =  3
9.0    86
1.0    80
4.0    76
8.0    76
2.0    75
6.0    74
0.0    73
5.0    69
7.0    68
3.0    56
Name: C, dtype: int64

[9.0, 1.0, 4.0, 4.0, 0.0, 6.0, 4.0]
The number of followers of 3 is 734
The number of followers derived from the observations is 7
Mode of most recent followers set is 4.0

 

A, B, and C are the positions. For example 308 is A=3, B=0, C=8.

under the last drawn (A B or C) is the list of 0-9 sorted by the most times appearing as a follower to A B or C.

the Name: etc is from using pandas data frames.

The recent followers represent the most recent 1% of the followers list. The mode is there to do a quick capture of trends of the small recent list. This was intended to use as a "quick pick" when pressed for time.

This pick based on mode is 6-5-4 for PA mid day. Night looks identical and calls for 3-8-5

I guess time will tell. Going to post these to predictions for the fun of it.

Working with Python has been a bit easier than trying monster excel formulas for sure. The current project gives me the entire list of followers for the last number drawn, including a count of the followers for that number, and the most recent 1% to allow a visual of recent trends. Then added a mode for the trends list to make a quick "pick".

Because I have always looked at the pick 3 as 3 separate games of 1 in 10, this lends itself to coding that is able to be used again.

The script has a function for pick 3 that has 3 of the same algorithm working on a different column of the .csv file (imported as a pandas data frame). This enables adding an additional algorithm to process different history files, so adding pick 4 and pick 5 should not take much time. Doing that tonight.

Paper play has no solid results yet, maybe later when I get to the laptop I can post a screen shot of what I am working with.

The 1% recent trends list seems to be the tweakable parameter. I have decided to try this route as the strictly excel based systems were getting too complex to maintain for daily games.

Anyone else using Python for exploring this hobby?

Got a Python program to automate counting followers for the last entry in the csv file, that is a time saver...

Picks for today in PA

3mid.  158

4mid.  4447

3eve.  175

4eve.  3794

Good to be back! Noticed I have been an LP member for 20 years.

After a number of years in the "I give up" gang, I realized that the work I was doing with lottery data was good practice with excel and I am starting up a new angle. In the interim I have learned to code in python, in fact just 8 classes to go (the long path, 1 at a time) to earn a Bachelor's degree in Computer Science.

Right now the current project is to generate predictions using Python code I wrote for time series analysis. I can import a spreadsheet (as a csv file) into a pandas dataframe and attempt fitting each ball in the PA pick 3 using SARIMA, then generating predictions for 7 days.

It is still a work in progress, but getting closer each day. The S in sarima means it looks for seasonal trends (something I though about but never implemented in excel).

Still treating the pick 3 as 3 separate draws of 1 in 10, and still not playing box, as a box means the code was wrong.

So many toys (algorithms) in the python stats and data mining tool box! After SARIMA, I will rework the code to use Markov chains.

Does not even matter if it works, the point is I am actively working with writing code.

No wins yet this year, stopped playing at all every few weeks out of frustration.

Newest system had some near wins, older system off by one or two a few times but nothing to cash in and I am just about out of ideas...

too late to throw into the "predictions" page, but the latest iteration of the probability system is showing these for tonight in PA;

P3: 7-4-6

P4: 7-3-5-2

P5: 5-4-2-5-5

Being disgusted with the near misses, I only went $1 straight on each. Not even bothering wasting money on box play because a hit there means I was wrong anyway and you end up spending twice as much. Still a few hours before draw time. If anyone plays them, good luck (and I'm sorry in advance because the system has not done anything since winning 5k in December 2018).

If anyone has any new ideas or rabbit holes to run down, I'd love to hear about it!

The lottery year was almost a complete flop. Middle of December used this crazy "triangulation by consensus" system and caught 5-3-4-3 straight on the pick 4 in PA. Had 2x.50st + 2x.50bx = $5,400

Could not get a follow up hit. No jackpot wins and a $290 .50st/.50bx on the pick 3 with the older ABS (ABsolute Subtraction) system.

Draws were so off following the win, I switched back to "paper play" just to track the errors without as much frustration.

Going to take another shot at the pick 3 today with the new system vs. the older ABS system.

Mid Day pick 3

TBC  5-6-5

ABS  7-4-2

Evening PA pick 3

TBC 5-0-6

ABS 4-8-2

On to 2019...

Going to try to use python to create a backtest for the new system, also importing the lottery draw csv files into the R language IDE for some genuine statistical analysis (and hopefully inference), finally going to use programming to yet again try and decode the seed scheme used by the daytime RNG based PA games...

Goal this year... a win on the p3 or p4 to help fund the ultimate goal, a win on the pick 5 ($50k for $1 straight)

 

Strange how both of my winning tickets for the pick 4 were for the 12/16/2018 drawing yet on the W-2G tax forms both are listed as 12/18/2018 for box 2;"Date Won"

Finally done updating the databases for PA pick 3 and pick 4 (not updated since 4/30/2017).

Made the sheets as described, took it to the next level by adding 2 more follower counts based on the last draw, the one before that, and the one before that... 3 data sets pointing to the same event from the same history. While still in the first few days of watching the draws (read that as paper play), PA mid day pick 4 was 5343, my sheet indicated 5343... oh well, wasn't playing for real anyway yet, but that would have been $2,500 on a .50 bet. Seriously, 2 draws into watching the "system" and there is a "hit" (don't like to count them as hits if I did not play them, but it is what it is).

Finished the pick 3 sheets 10 minutes ago... will be watching them also.

Sometimes the systems I come up with are just too difficult to accurately back test, this is one such system... so now I must wait and see how long it takes to get another "win" so I have an idea of the gap before deciding to play around with actual tickets. took literally decades for me to figure out how to do what I had envisioned with a spreadsheet, took 20 minutes to convert from the pick 4 to the pick 3, will be similarly quick for eventual conversion to the pick 5 (got to get the columns right in all the counting formulae.) when I finally finish adding the missing draws to that DB.

Since I like to have a dumb name for each system, I dub this one "Triangulation by consensus", and though I may not yet know how to interpret what I am looking at, I have already begun the process of recording the picks vs results for gathering error data. Hopefully can figure out how to get an error correction that makes sense... last system left a lot to be desired in that department, even when squaring the errors and dividing by 2 to eliminate negatives.

Back to the drawing board after mixed (mostly losing) results with the previous system in the pick 4.

Difference since 2016... Took some math courses at an online university (algebra with trigonometry, calculus, statistics, discrete math, and currently going through statistical inference). Going for a bucket list bachelor's degree in computer science. Will take 6 more years... hanging on to a 3.3GPA and lerning a great deal. This is my last math class that the school even offers, so it's all cores and "fluff" from here on out... 9 more classes after this one to an Associate's Degree.

Also dabbling more seriously in programming. 1 class in Python and 2 in Java so far. Also learning Android programming with being awarded a Google scholarship to attend Udacity for the Android Basics nanodegree, which I have recently completed. Currently in a free class for Machine Learning with Python and Pytorch.

plus full time work, plus 5 grandkids (and one on the way)... no rest. AND just hit 50 (was not looking forward to that one).

Next up is a look at the numbers that follow other numbers, basically going to take each number and count the number of times that each of 0-9 followed it. Then, convert these sums into a probability distribution and see what I can do with that. After studying such fun stuff as linear/logistic regression, batch gradient descent, bayesian inference and deep/shallow learning, I think that help might be available but we must figure out what question to have the neural network answer... Rather than have it directly predict a umber, it might be best to have it predict the results of some statistic of the number, like the one described above.

My database for pickN in PA for the 2-3-4&5 are about 18 months behind, so that will be the first phase... And after knowing all this new stuff, there still seems no better tool to visualize what is going on with the lottery than a spreadsheet. (Although I have been quite intrigued by what the language R is capable of, will check out some population stats when I get the database caught up... too bad lazy PA does not have a decent API to get the latest draws from.

So in the never ending quest for a winning system, it's always ending up back to excel.

The challenge for the new system was to keep it cheap enough to stick with, hence the new strategy

1. Alternate between Pick 3 Pick 4 and Pick 5 (different game per week)
2. Play one straight and one box for day and night
3. only play on days I work (4 per week)

Given these parameters, play on the pick 3 and pick 4 is $2 per day and $4 when dealing with a pick5 week.

As for the system, something stupid and simple that didn't even backtest all that well

the ABStraction system (coincidental at best)

Using a simple ABS() formula in excel, subtraction can be commutative like addition. I developed a "mask" that you simply subtract from the last number drawn in each position and play the resulting number. Updating the mask based on backtests means you can just keep a copy of the mask and not need to be constantly at the spreadsheet, weekly updating suffices.

for example

last PA pick3 mid-day was 8-8-3

MASK is 8-7-3

8-8=0     8-7=1   8-8=0, play 0-1-0 if there was a higher top number it doesn't matter, just use ABS() on the answer to remove negative numbers.

for example tonight's pick3

last pick 3 was 0-5-0

Mask is 7-9-4

ABS(0-7) = 7

ABS(5-9) = 4

ABS(0-4) = 4

7-4-4 for tonight.

play should be 0-1-0 for tomorrow mid day... not playing because it's a day off so... cheap system!

results thus far

Pick3   3 box hits ($40 each) / 0 straight hits

Pick4   0 box hits / 0 straight hits

pick5   1 box hit (2 pairs, $1,700) / 0 straight hits

 

So with an annual projected cost of $560, the $1,820 earned so far makes it a personal success, even subtracting the $425 due in taxes on the pick 5 hit... it just doesn't backtest favorably. 

The only system I ever had that had consistent positive results was erased years ago when Microsoft chose to wipe out VB compatibility between 6 and .net  Now they offer a community edition of visual studio for free, including fully functional VB and C#, so maybe I can brush up on programming and revisit the idea, but for now this system is cheap and entertaining.

results may vary, entertainment value only... but you never know.

the MASK values were obtained through brute force backtesting, trying 0 thru 9 in place on excel, so anyone with excel can recreate the sheet.

for each position, you get a mask number therefore it's not good for trapping or gathering groups of numbers, and on most days it's almost a guide for what NOT to play, but it's cheap and so far avoiding straight hits, it does well for boxes, particularly on the pick 3. It does favor some numbers, but none are out of play, so it won't help reduce either.

Still fiddling with the spreadsheets, but thought I would try a test fire number for the week in PA

1-1-8-1 on the pick 4 (thursday eve thru tomorrow mid)

frustrating on 9/6 when 1-8-2-1 came out, but quite a slap in the face today when they drew 1-1-8-5...

3 numbers IN POSITION, actually played the ticket, fourth number off... I am remembering why i usually stick to "paper play"...

oh well, got tonight and tomorrow mid yet, then back to paper, obviously system not quite ready for prime time.

same system likes 5-1-4-0 for next week... gotta decide if it's worth the couple bucks to ride the rollercoaster again or just re-tool the sheet and watch from the sidelines again.

Sometimes it gets discouraging to envision what you want to do on a sheet of paper, craft it into an excel spreadsheet and realize the information that you wanted is indeed presented correctly BUT provides no help whatsoever when the numbers are drawn...

I knew straight statistics wouldn't be the holy grail, otherwise everybody would win.

So what I have left is my original design goals...

1. treat each digit separately (pick 3/4)
2. make a unique pick for each draw
3. make ONLY one pick for each draw

Sometimes you get tempted to toy with settings to "enhance" backtest results, knowing that it won't propagate forward... I just saved myself quite a significant amount of time by realizing and stopping this in my current endeavor.

Since the project began, i had managed to avoid any straight results, but picked up $440 in box results because it was cheap to add them, but in the end, box was wrong... but funded the system thus far. 

So i sit with my excel skills getting slowly sharper, but positive results still missing. BUT unlike last time, i refuse to allow myself to drop my experimentation and computer skills due to lack of results. The current framework is good, I must wait for the next idea or inspiration, but the framework is there... updating draws until then.

So, i believe the statistics are a part of the puzzle, but after exhaustive attempts, I now know they are not the ONLY part...

but what next?

Coming up with spreadsheet ideas, sometimes you have a ton at once, sometimes not a new idea for weeks... currently in the latter.

Updating the draw databases I have created without even pulling a pick until after the draw... wanting to try a few games, but heeding the backtest results that are red flagging the idea of live play at this time. Thinking about taking advantage of the writer's block and learning something useful like Java programming... also cheap! maybe helpful with this project as well...

Given the breakdown of my current project, picking 1 in 10 consecutive times... there is a standard 10% chance in a random pick, and by studying history I can only get near 12%. time to keep looking...

So, what are the metrics required for building a system? how could you use past draw information to pick future draws?

Fundamental questions to be sure, and considering what we try to do everyday, ones that need some attention.

I have come up with 4 tools to add to a draw profile, for each position in a pick3...

1.Count... show how many times 0-9 followed the currently drawn number throughout your selected draw history.

2.Day Of Month... based on the 1-31 range, shows 0-9 for the last draw on THAT day of the month, a filter... a subset of the data above.

3.Day Of Week... based on 1-7, so i can see what followed a 9, for example, every time a 9 came up on a tuesday...

4.Simple follower... if a 6 is picked tonight, and a 3 was picked last night, how do 0-9 look when the 3-6 chain is previously drawn.

 

now the question... which is more important? the one with the highest count? the one with the highest opportunities for a match?

I am developing a backtest that will check all 4 tools against the draw history to see if any could have been used for a winning system... found some glimmer of hope on the midday drawing using the day of the month, using just the MAX count it would have paid 5 straight hits in 2,200 draws, which on $1 st. would leave a profit of $300, currently... 

So much to do before even paper testing can begin...

So, what tools are in your system? (Notice I was not counting Even/Odd, Hot/Cold, Low/High... filters not used in a system where the end result is just one number).

Haven't had a post in the forum since 2009!!!

Haven't run an excel spreadsheet since then either... anyone having any luck?

currently getting acquainted with the commands in excel 2007 (old laptop)... looking to finally code draw profile info that will now be possible with the COUNTIFS() function... so far just a concept, not even paper play yet... the brief summary

1. reduce pick 3 to consecutive games of 1 in 10
2. fine tune it
3. scale it up to QUINTO (skipping the pick4)