hypersoniq's Blog

Here is the output... (5-8-3 mid day and 8-5-0 evening). note how the list lengths equal the last draw result... Auto ranging! Now let's see if it might work...

===================================================================================
Pick 3 Mid Day Followers
===================================================================================

The number of followers in the RECENT lists: 5, total followers evaluated: 35
Mode of most recent followers sets combined (The pick!) is >>> 5.0 <<<

-------------------------------------------------------------

The number of followers in the RECENT lists: 2, total followers evaluated: 14
Mode of most recent followers sets combined (The pick!) is >>> 8.0 <<<

-------------------------------------------------------------

The number of followers in the RECENT lists: 3, total followers evaluated: 21
Mode of most recent followers sets combined (The pick!) is >>> 3.0 <<<

-------------------------------------------------------------

===================================================================================
Pick 3 Evening Followers
===================================================================================

The number of followers in the RECENT lists: 8, total followers evaluated: 56
Mode of most recent followers sets combined (The pick!) is >>> 8.0 <<<

-------------------------------------------------------------

The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 5.0 <<<

-------------------------------------------------------------

The number of followers in the RECENT lists: 4, total followers evaluated: 28
Mode of most recent followers sets combined (The pick!) is >>> 0.0 <<<

-------------------------------------------------------------

====================================================================

The current follower system has a setting for list length that can be unique to each game, but the same across positions. That is what I overlooked. The ability to set the list length for each position... I will remedy that!

Each position in pick 3 (and beyond) has it's own history, and might give better results if it had it's own list length. I will set up the unique variables to pass these in during the function call as parameters. I will determine in the short term here which lengths to use by solving tonight's draw BEFORE updating the csv draw files. Though it has the potential for over fitting, the limits will be to choose the lowest list length that contains the winning number... even if it was not the one picked. May take longer than one night to implement, but I think it is the next logical step.

I have already started the script that will perform the back test, begun by suppressing the verbose output and limiting it to list length (which will also be suppressed later) and the pick.

Going to try starting at between 2 and 3 years of seed data, for example pick 3 from the start to dec. 31st 1979 and begin that back test from January 1st, 1980, as the past draws are far back enough to allow follower data, and that information would have been a priori.

A statistics based system might not be glamorous or as fun as picking 100 numbers for all states, but if it works at an interval to hit enough to be profitable, then it will be worth the time, and for the jackpot games implementation, that only needs to work once.

While calibrating the dailies (pick 3, 4 & 5) on paper, I ran the mega millions data for 7 different settings. It picked several mega balls, but the most consistent one was 25, the one drawn last night,  white balls could only manage one. This might not be the tool for the white balls, but it does seem useful in narrowing down a good Mega Ball pick. Testing tonight on the power ball.

I played 7 lines of MM last night (total cost $21, with megaplier) and brought in $42.

Letting that $42 roll with similar test on the power ball tonight. The difference being the cost of the test... as I keep separate data for the power ball double play that they pick in PA. The $42 will require an additional $14. Nothing ventured, nothing gained. The results from last night let me pick a setting for MM and for all of the un-played draw games. The one annoying anomaly... again the mid day winner (straight) was among the evening picks... twice in 2 weeks.

After this test to dial in the power ball, I will have all the list lengths set and go back to 1 pick per draw, no guarantee, but I will honestly be able to say it is my best guess.

Working out the most affordable play strategy, then a proper live test can be conducted (with LP predictions to record progress)

Made a copy of the script to start modifying output for the back test. Not going to be easy, but will be worth the knowledge gained in manipulating 2 CSV files in the same script at the same time.

The ultimate goal is one pick for one game. There are really 3 parameters I can adjust...

1. The number of recent followers in each list

2. The number of lists

3. The skip length of each list

The day for night thing was an observation when the pick for day came out straight for night, but that adds cost and defeats the original purpose of the system.

The list length is the easiest to adjust. Perhaps the next move should be to try several list lengths at once to see which one got closest to the last draw before updating the databases today. That sounds like the plan. I expect that each game may have it's own list length setting.

Also making a copy of the script, suppressing all output but the picks and getting the new script ready to process the back test.

One test for the new script is setting the most recent list length to one... accomplishes an expected behavior and a fall back one...

1. Expected behavior is to return the mode, or most frequent of the 7 individual followers.

2. Fallback behavior... if all 7 are different, choose the first follower (the most recent).

No code changes were required, just passing each function the value of 1 for number of followers. It is looking at

Direct follower

Skip 1 follower

Skip 2 follower

Skip 7 follower

Skip 14 follower

Skip 21 follower

Skip 28 follower

So it always has 7 numbers to work with.

Worth a test on tomorrow's PA pick 3...

Mid day best guess... 289

Eve best guess... 965

Based on recent observations, going to play each pick for day and night.

Mixed result... system pick was 0 matches, but the last QP line had 5 of 6, and the 6th winning number was above on qp line 2...

So the system did not work for the win, but I would not have played that ticket if I wasn't testing the system.

Total win was $1,025

The other anomaly last night was on the pick 3, the number that my system gave for the mid day came out straight at night... did not play day for night because of the added cost, but I couldn't help but notice that.

Might be worth keeping an eye on for the rest if the week. Playing each pick for both doubles the play cost to $4/day and shortens the profitability window to a win within 125 draws.

Still planning the back test parameters, possibly limiting the starting pool to 3 years as the other games were not around as long as the pick 3.

It did not escape me that I was 1 number away from 1.6 million last night. But since it was on a QP line, it does not give the system development any real motivational boost. That motivation comes from the challenge of applying unit test best practices to the back test phase.

It is all about gaps. The gaps between wins.

What you can tell right away...

A maximum gap (longest stretch between wins) of 501 days or more indicates a system that is not profitable.

The average gap has no value.

The minimum gap is the number of days you can confidently STOP playing. If you don't play every draw, your expense in chasing the next win is reduced! If there is an anomaly like a back to back win, it can most likely be discounted as an outlier. By looking at the gap data, there is a statistics formula to determine outliers.

So the use is...

Max gap determines profitability... is the system worth pursuing?

The min gap determines when to NOT play to save on the play budget.

I am still on the part of my system where I can get picks, but have not finished the back test yet. I may play on occasion to try and get a win, but I won't have this info immediately. 

There are no guarantees, but this idea helps reduce the play expense, thereby maximizing profitability.

The initial scenario that comes to mind when envisioning a proper backtest for this follower system would be to 

1. Pick a point in history where the test should start. A follower system should have some data to work with. Perhaps the number of draws in the pick 5, around 5k. There are 16k draws on the evening pick 3, so that seems about right.

2. Run the program with only that history, record the pick.

3. Fetch the next result from the full history file.

4. Compare the pick with that result, if a match record the date and number.

5. Re-run the script, repeat until the history file is fully loaded and there are no more draws to add. 

Assuming the cyclic rate hovers around 1 second, there will be 11,000 ish iterations, so run time can be predicted to be around 3 hours.

I should only need to do this for the pick 3 games, then I can say I gave my all in the due diligence department, plus it will be experience in writing unit tests... a real opportunity to use software engineering best practices!

After that, the back test can be used to check settings at 3 hours a shot. Onward!

As promised, the output of my Python script for the drawings of the PA Pick 3 (Mid day and evening). I will also post the predictions. There is a great deal of info, but the pocks are easy to see, surrounded by >>> <<<. the picks are as follows (For Tuesday, August 22nd, 2023) ; 

MidDay: 4-8-9

Evening: 0-2-0

====================== RESTART: C:\Python39\superSeven.py ======================
Daily Lottery Follower Frequency Analyzer
Super Seven Edition!
Software by HyperSoniQ
 
===================================================================================
Pick 3 Mid Day Followers
===================================================================================
last drawn A =  4 , secondary =  9 , tertiary =  3
Skip Seven =  2 , Skip Fourteen =  6 , Skip Twenty One =  8
Skip Twenty Eight =  2

Primary Followers :  [7.0, 3.0, 4.0, 8.0, 8.0, 3.0, 6.0, 8.0, 0.0]
Secondary Followers :  [3.0, 7.0, 4.0, 4.0, 8.0, 1.0, 8.0, 2.0, 2.0]
Tertiary Followers :  [2.0, 6.0, 7.0, 2.0, 6.0, 4.0, 4.0, 5.0, 0.0]
Skip 7 Followers :  [9.0, 8.0, 4.0, 0.0, 5.0, 6.0, 3.0, 1.0, 9.0]
Skip 14 Followers :  [6.0, 7.0, 4.0, 6.0, 7.0, 7.0, 5.0, 3.0, 1.0]
Skip 21 Followers :  [4.0, 2.0, 5.0, 6.0, 4.0, 7.0, 7.0, 4.0, 9.0]
Skip 28 Followers :  [8.0, 8.0, 7.0, 3.0, 8.0, 0.0, 0.0, 3.0, 4.0]
The number of followers of 4 is 768
The number of 1 skip followers of 9 is 750
The number of 2 skip followers of 3 is 748
The number of 7 skip followers of 2 is 742
The number of 14 skip followers of 6 is 742
The number of 21 skip followers of 8 is 725
The number of 28 skip followers of 2 is 740
The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 4.0 <<<

-------------------------------------------------------------

last drawn B =  8 , secondary =  1 , tertiary =  8
Skip Seven =  1 , Skip Fourteen =  1 , Skip Twenty One =  8
Skip Twenty Eight =  5

Primary Followers :  [1.0, 8.0, 0.0, 7.0, 7.0, 5.0, 1.0, 0.0, 5.0]
Secondary Followers :  [7.0, 0.0, 8.0, 1.0, 8.0, 3.0, 1.0, 2.0, 8.0]
Tertiary Followers :  [8.0, 0.0, 1.0, 8.0, 3.0, 1.0, 4.0, 9.0, 5.0]
Skip 7 Followers :  [1.0, 6.0, 7.0, 1.0, 0.0, 0.0, 5.0, 8.0, 6.0]
Skip 14 Followers :  [8.0, 1.0, 7.0, 0.0, 8.0, 5.0, 6.0, 8.0, 2.0]
Skip 21 Followers :  [6.0, 8.0, 0.0, 0.0, 0.0, 3.0, 5.0, 2.0, 6.0]
Skip 28 Followers :  [1.0, 6.0, 1.0, 3.0, 8.0, 0.0, 3.0, 5.0, 6.0]
The number of followers of 8 is 746
The number of 1 skip followers of 1 is 760
The number of 2 skip followers of 8 is 745
The number of 7 skip followers of 1 is 759
The number of 14 skip followers of 1 is 758
The number of 21 skip followers of 8 is 741
The number of 28 skip followers of 5 is 726
The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 8.0 <<<

-------------------------------------------------------------

last drawn C =  2 , secondary =  9 , tertiary =  2
Skip Seven =  8 , Skip Fourteen =  9 , Skip Twenty One =  3
Skip Twenty Eight =  8

Primary Followers :  [9.0, 4.0, 3.0, 1.0, 9.0, 9.0, 9.0, 6.0, 9.0]
Secondary Followers :  [4.0, 3.0, 5.0, 4.0, 6.0, 3.0, 8.0, 8.0, 5.0]
Tertiary Followers :  [2.0, 2.0, 6.0, 3.0, 4.0, 3.0, 0.0, 2.0, 0.0]
Skip 7 Followers :  [2.0, 3.0, 9.0, 0.0, 3.0, 3.0, 9.0, 4.0, 9.0]
Skip 14 Followers :  [0.0, 5.0, 4.0, 1.0, 9.0, 2.0, 3.0, 2.0, 8.0]
Skip 21 Followers :  [9.0, 2.0, 0.0, 9.0, 6.0, 5.0, 9.0, 2.0, 1.0]
Skip 28 Followers :  [4.0, 4.0, 2.0, 0.0, 0.0, 1.0, 3.0, 8.0, 5.0]
The number of followers of 2 is 758
The number of 1 skip followers of 9 is 778
The number of 2 skip followers of 2 is 757
The number of 7 skip followers of 8 is 772
The number of 14 skip followers of 9 is 777
The number of 21 skip followers of 3 is 731
The number of 28 skip followers of 8 is 770
The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 9.0 <<<

-------------------------------------------------------------

===================================================================================
Pick 3 Evening Followers
===================================================================================
last drawn A =  0 , secondary =  6 , tertiary =  7
Skip Seven =  9 , Skip Fourteen =  9 , Skip Twenty One =  0
Skip Twenty Eight =  6

Primary Followers :  [8.0, 0.0, 8.0, 1.0, 5.0, 6.0, 3.0, 8.0, 6.0]
Secondary Followers :  [7.0, 9.0, 0.0, 5.0, 3.0, 8.0, 1.0, 4.0, 0.0]
Tertiary Followers :  [0.0, 9.0, 5.0, 5.0, 0.0, 3.0, 5.0, 7.0, 7.0]
Skip 7 Followers :  [9.0, 6.0, 3.0, 6.0, 8.0, 2.0, 6.0, 0.0, 8.0]
Skip 14 Followers :  [6.0, 9.0, 0.0, 7.0, 8.0, 4.0, 4.0, 1.0, 3.0]
Skip 21 Followers :  [0.0, 6.0, 4.0, 5.0, 9.0, 8.0, 5.0, 1.0, 4.0]
Skip 28 Followers :  [0.0, 9.0, 1.0, 9.0, 6.0, 5.0, 4.0, 2.0, 7.0]
The number of followers of 0 is 1693
The number of 1 skip followers of 6 is 1632
The number of 2 skip followers of 7 is 1686
The number of 7 skip followers of 9 is 1550
The number of 14 skip followers of 9 is 1549
The number of 21 skip followers of 0 is 1692
The number of 28 skip followers of 6 is 1628
The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 0.0 <<<

-------------------------------------------------------------

last drawn B =  5 , secondary =  6 , tertiary =  4
Skip Seven =  4 , Skip Fourteen =  7 , Skip Twenty One =  9
Skip Twenty Eight =  3

Primary Followers :  [2.0, 4.0, 6.0, 0.0, 8.0, 1.0, 8.0, 7.0, 7.0]
Secondary Followers :  [2.0, 1.0, 1.0, 8.0, 4.0, 8.0, 5.0, 2.0, 8.0]
Tertiary Followers :  [4.0, 2.0, 6.0, 7.0, 2.0, 7.0, 2.0, 4.0, 0.0]
Skip 7 Followers :  [9.0, 5.0, 0.0, 9.0, 2.0, 9.0, 2.0, 0.0, 5.0]
Skip 14 Followers :  [9.0, 4.0, 5.0, 1.0, 9.0, 4.0, 5.0, 0.0, 8.0]
Skip 21 Followers :  [4.0, 0.0, 0.0, 7.0, 8.0, 2.0, 9.0, 1.0, 3.0]
Skip 28 Followers :  [7.0, 1.0, 6.0, 5.0, 2.0, 0.0, 9.0, 2.0, 1.0]
The number of followers of 5 is 1661
The number of 1 skip followers of 6 is 1638
The number of 2 skip followers of 4 is 1614
The number of 7 skip followers of 4 is 1612
The number of 14 skip followers of 7 is 1641
The number of 21 skip followers of 9 is 1613
The number of 28 skip followers of 3 is 1594
The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 2.0 <<<

-------------------------------------------------------------

last drawn C =  2 , secondary =  6 , tertiary =  0
Skip Seven =  6 , Skip Fourteen =  2 , Skip Twenty One =  4
Skip Twenty Eight =  3

Primary Followers :  [0.0, 2.0, 5.0, 4.0, 8.0, 2.0, 7.0, 2.0, 5.0]
Secondary Followers :  [0.0, 6.0, 9.0, 3.0, 6.0, 4.0, 4.0, 5.0, 6.0]
Tertiary Followers :  [6.0, 2.0, 4.0, 4.0, 6.0, 5.0, 5.0, 2.0, 1.0]
Skip 7 Followers :  [0.0, 0.0, 2.0, 8.0, 8.0, 4.0, 0.0, 0.0, 4.0]
Skip 14 Followers :  [2.0, 8.0, 2.0, 4.0, 5.0, 8.0, 2.0, 4.0, 0.0]
Skip 21 Followers :  [0.0, 2.0, 3.0, 8.0, 5.0, 6.0, 0.0, 0.0, 5.0]
Skip 28 Followers :  [2.0, 1.0, 4.0, 0.0, 0.0, 1.0, 0.0, 9.0, 2.0]
The number of followers of 2 is 1653
The number of 1 skip followers of 6 is 1678
The number of 2 skip followers of 0 is 1592
The number of 7 skip followers of 6 is 1676
The number of 14 skip followers of 2 is 1652
The number of 21 skip followers of 4 is 1590
The number of 28 skip followers of 3 is 1611
The number of followers in the RECENT lists: 9, total followers evaluated: 63
Mode of most recent followers sets combined (The pick!) is >>> 0.0 <<<

-------------------------------------------------------------

====================================================================
Program has completed!
Runtime = 0.003891619708802965
Soniq BOOM!

The process will be to generate picks for both the mid day and evening PA pick 3. This is paper play (no tickets bought) and the new draws will be the targets. Adjusting the number of recent draws to capture with the goal being the lowest number of draws needed to pick correctly.

Box hits do not count, as if the numbers come up boxed, the result was still wrong.

The draw databases will NOT be updated until the system is calibrated for today's results. This process may not even pick the winner, it just has to be such that the winning numbers were not excluded because the list length is too short or over fitted by having the list too long. This could take days. Once it looks like the list lengths are satisfactory, the first live fire test will happen.

The "system" to accompany these results are as follows. $1 straight on the mid and evening PA Pick 3's will be played. No boxes. I will post the output of the live fire test (and post in the prediction section) so anyone can see the output. The readability mods made it much easier to follow, including labels for the lists and picks are output as >>> n <<<. That makes it easy to quickly spot the picks amid the other data.

The initial script was designed to handle the pick 3, pick 4 and pick 5... but the addition of a module to process the PA Match 6 can be recycled on the Mega Millions and Power Ball.

The hardest part of this project will be creating a back test, but perhaps a modified version that increments the exposed draw history and outputs picks to a csv file can be written, then pasted to a database copy. The back test may take quite some time, but is needed to estimate hit frequency (or lack thereof).

I have made many failed systems in Excel, but retained the skills learned. The same is true of Python coding, so even if it does not work it is still a win.

If you play the pick 3, and you play one number a day using whatever system, you need to win once every 16.43 months to break even. I have yet to find a single pick system that can back test to this level.

If you play both the mid day and evening games, you need a win on either within 250 days to break even.

Profitability requires an even higher win rate.

This is why I stick to 1 pick per game. Regardless of the game.

The follower script already tracks the following:

Direct followers

Skip 1 followers

Skip 2 followers

To this list I am preparing to add:

Skip 7 followers

Skip 14 followers

Skip 21 followers

Skip 28 followers.

Why? More perspective. With 7 different events all pointing to the next draw, maybe it will have a higher chance of accuracy.

I still encounter situations where either the winning number was eliminated from the pool of possible picks or the most frequent of the respective lists gave a greater weight to a number that was not drawn. I will shorten the length of the recent individual lists and look at it from 7 different perspectives, combining these to generate the one "best guess" based on previous draw history.

It still presents my ideal system which does the following:

Treats pick N games as N sequential games of 1 in 10

Uses draw history

Updates and runs quickly

Uses Python code rather than the longer process of updating spreadsheets

Generates one single pick definitively, which reflects my cheap mantra.

May not work ever, but represents my single best guess based on my years of research (on and off, mostly off)

Keeps the coding skills sharp.

 

I will temporarily suppress the full history counts to make it more readable as I tweak the settings, mostly of recent capture length. This is simply done by commenting out the print lines that display the full counts.

I may also display the individual picks while testing for the purpose of determining which component contributes the winning pick while paper play testing the pick 3, but it is the composite pick that is the sole target of these experiments. This will be my first expedition into feature weighting... a positive step to understand the mechanics of machine learning, which will be the eventual goal.

Exciting to be working on 2 of my favorite hobbies at the same time... Python coding and the Lottery!

I have not tracked PB or MM in a long time. I found the fastest way to get a spreadsheet for each game was through the PA lottery crappy website 1 year at a time. Copy/paste (special) then clean and sort. Took a little under an hour to create spreadsheets with the current matrix in both games. Too bad PA doesn't have an API (application programming interface) so the process could be automated with a decent script. Still sharp on the spreadsheet skills though! The current matrix data on USA Mega is a great resource, thanks to the poster who shared it on the Jackpot Games forum!

Just out of curiosity, how do you gather data?

Here is a sample of the output, only put the first digit here of the PA pick 3 mid day as you will see it prints out a great deal of info...

 

Daily Lottery Follower Frequency Analyzer
Triple Play Triangulation Edition!
Software by HyperSoniQ
 
====================================================================
Pick 3 Mid Day Followers
last drawn A =  8 , secondary =  0 , tertiary =  8
7.0    96
8.0    85
0.0    81
5.0    73
4.0    71
3.0    70
1.0    69
9.0    63
6.0    63
2.0    57
Name: A, dtype: int64

3.0    89
0.0    87
2.0    86
7.0    81
4.0    79
8.0    75
6.0    71
1.0    70
9.0    67
5.0    56
Name: A, dtype: int64

8.0    84
3.0    78
0.0    77
1.0    77
4.0    75
6.0    75
5.0    70
7.0    68
2.0    65
9.0    58
Name: A, dtype: int64

Primary Followers :  [0.0, 8.0, 6.0, 8.0, 2.0, 0.0, 7.0, 5.0, 3.0, 5.0, 2.0, 8.0, 9.0, 0.0, 6.0, 8.0, 8.0, 7.0, 3.0, 6.0, 7.0, 8.0, 9.0]
Secondary Followers :  [8.0, 1.0, 6.0, 7.0, 8.0, 0.0, 3.0, 6.0, 4.0, 0.0, 9.0, 6.0, 4.0, 2.0, 8.0, 0.0, 3.0, 4.0, 0.0, 8.0, 8.0, 3.0, 1.0]
Tertiary Followers :  [nan, 8.0, 7.0, 1.0, 5.0, 1.0, 6.0, 7.0, 5.0, 7.0, 1.0, 6.0, 8.0, 6.0, 0.0, 4.0, 6.0, 7.0, 8.0, 2.0, 6.0, 8.0, 3.0]
The number of followers of 8 is 729
The number of 1 skip followers of 0 is 762
The number of 2 skip followers of 8 is 729
The number of followers in the RECENT lists: 23, total followers evaluated: 69
Mode of most recent followers sets combined (The pick!) is 8.0

 

nan is at the head of the list, so it sometimes pops up in the tertiary followers (skip 2s)

Noticed a MAJOR bug since the last version... the lists were backwards! oldest to newest... a list.reverse() fixed that error AND I verified the data by checking the csv to match what the output of followers (and skip1s, skip2s) was in history. Now the data is validated and the system operates correctly. But does it work? time will tell...

Now that I have the followers script functional (it has not picked a winner yet, but it has only been run a handful of times) it is time to expand on the idea I had for followers last manifested in a spreadsheet... "triangulation"

This is where you observe the followers of the last drawn number, also the 1 skip follower of the draw before last and the 2 skip follower of the draw before that... like triangulating the best pick from 3 different vantage points.

I had this in a spreadsheet that took hours to update for all of the games. What is new is the ease of only updating the draws in a csv file and the most recent followers component.

Hopefully after tomorrow, I will have the ability to analyze all 3 follower lists, create the short list of recents, then combine these 3 lists to get one pick.

Then the settings can be altered for the length of the recent pick list. That will be accomplished by not entering the last draw until the settings produce the winner. By only tweaking the list length and maybe the skip distance I do not run the risk of data starvation (not enough, where the winning number was eliminated) or overfitting (where the data is optimized for just one draw).

Given how long the old spreadsheet took to update and the fact that I did not have a "most recent" list, I hope that I can at least explore all of the possibilities of this concept.

Even if the run time will be tripled, the script runs in less than a second now.

After the new mods are in place and functional, I start looking at graphs generated using the data such as those from the matplotlib or seaborn libraries... really enjoying Python these days!