hypersoniq's Blog

Pulling the plug on the start day test. Data was mostly the same stepping through 7 draws.

Still live testing the pick 3, just not shuffling start days.

There is the added change each week in interpreting the results, so rather than go for constant play, I will be instead running if I have identified an area of interest to focus upon and pursue an attempt. This will ultimately lead to less frequent attempts and more time at the phase 2 drawing board... which is generally how I like things.

I do realize that my observations can be prone to bias, and therefore care is being taken at each step to avoid it. It feels like I am onto something, but again I am not the best judge. Time needed to study the results is also a factor in discontinuing play on any type of schedule. While I seek a clearer interpretation, the live game waits... when I am ready to play, it is back in with the same budget restrictions and overall goals.

I am asking a great deal of the coincidence machine, as the initial goal is a hit both day and night, and I know it will fail to deliver at a much higher rate than it might work... but that is the fun... the challenge.

While exploring my most recent ideas, I stumbled upon what I will be trying sometime before the weekend, which is to run the script as usual, then inject another run with a different model size. So, 150 as usual, then 143... why? The future will show which ones are to be drawn and which ones will not... BUT we know which ones will drop off between the week window! 1/3 of the future state of the distributions can be known ahead of time... using the 2 lists to find out which numbers will still be likely to hold neutral status throughout the window can hopefully be the indicator of which number to pick in a tie. There still remains 2/3 of the final state unknown... but before this observation it was fully unknown... I will take the 1/3 and run with it.

Still stumbling through code for the visualization, but this gives me something to work with now. Standard deviations will change, quartiles will shift, the data comes alive again!

As per usual, the tests so far have been a flop.

5 weeks remain of the test, but days may be skipped between.

Having viewed the 7 step results, there is an idea brewing...

Next pick, I will run the standard script, then change one parameter and run it again. The final distribution list of each will help narrow down what to play, but unsure as to how much help it will be...

I don't know which numbers will be picked, I don't know which numbers will not be picked... but I DO know which numbers will drop off of the list in a week...

With the pick 5 test I am $63 in, $70 to go.

$133 total expense

$0 in winnings

One p3 straight or 4 $40 box hits will make it a successful test. Total of 28 draws mid and 28 draws eve to go...

With the upcoming 7 day step modification, it might be a good time to get a visual look at the data.

The hard part is knowing what to look for...

Over 7 days of steps, each step brings in 1 new combo into the model, and loses the first combo. This means the numbers have 3 possible actions...

Increase by 1, stay the same, or decrease by 1.

So for a first attempt, I will be looking at using frequency to populate the Y axis, putting a line across the graph at the expectancy, a blue line below by 1 standard deviation and a red line above the expectancy by 1 standard deviation. This will define the neutral range.

For the X axis, it will be the steps.

What we will have is each number being plotted by frequency at each step.

Now, that could get to be a busy graph with 10 different plots on it, so I want to incorporate a list of check boxes, default UNchecked, for each digit 0 through 9.

I am hoping to see what happens with regard to frequency at the draw level, rather than the entire week to week.

I may have to re arrange the output so that the standard deviation and the quartiles appear alongside one column's worth of distribution data.  Such that it looks like 

SD, Q1,Q2,Q3,0 freq, 0 percent, ... 9 freq, 9 percent. Then this data will need to be parsed to either skip rows to ensure column 1 data lines up or create a separate output file for each column so only column A data appears with column A, etc.

So we run, then read column A data, then read column B data and finally read column C data. Once loaded, the graph and check boxes can be generated.

Because the program screen output will be unchanged, I can use the graph to study numbers of interest, see how several numbers compare (deciding between ties in frequency) or even see the whole cluster of a graph with all 10 digits present.

Since frequency is being used, ranging is easy for Y, it is 0 to 30, with 15 being the center (and the expectancy). X is simply 1 to 7 for the steps.

Getting the correct data written should be as easy as properly formatting one writerow() command per column pass.

Visualization will be a clean sheet separate script, so I can cross reference data in the format I have gotten used to reading with the new graphing function. I will keep in mind the flexibility of the original concept by writing the visualizer to accept a variable number of input files so the same program that can process the pick 3 can also process the pick 5.

In retrospect I should have done this BEFORE trying a live test, but ideas evolve... the live test might be over before this idea is functional.

Time will tell...

I knew a day or 2 would help with figuring out the next part.

Reconfiguring the program to show the results for each of 7 days is going to be simple.

1. Set the window to 0.

I am not looking to classify the next draws, just look at the resultant statistics from the model.

2. Set the offset to 7, then put the function in a loop with offset decrementing by 1 with each pass.

3. Call the function with the new offset.

This should show the distribution over 7 draws, step by step. This will help show how the statistics move in the middle of that 7 day window, how the distribution changes. Hopefully highlight exactly what I would be looking for in the stats to help with the next pick.

The work schedule has me on a 5 day stretch starting tomorrow, so it could be a few extra days until the opportunity presents itself to make a pick anyway, so may as well study the mechanics a bit closer.

This will be a branch of the original script, therefore it will get a new name and version number (V1), and will be used in conjunction with the other script to help pick a combo.

While it is nice to just run the script and study the resultant data, I think it would be a better experience if I were to create a full Python application.

For readability, the current script stores information in lists so that I can get a 3 column display output at the end. If I instead use a grid based system, they can populate directly after each column processes, allowing for the following...

1. I can sort the distributions descending!

2. I can change the background color of the cells in the distribution, red for Hot, Blue for cold, leaving the neutral with a white background.

3. I can use the same per column approach, where the grid layout would be dependent upon the number of columns of data in the input file, providing a 3x3 grid for output with pick 3 data, or a 5x3 grid for pick 5 data.

4. User controls can appear below the grid

5. The introduction of visualization!

This will make the data easier to read, and allow the development of another feature... step sequencing... taking the results from the 150 draws and stepping one draw at a time to watch the changes. Processing in 7 draw steps makes a big difference in the output, maybe a single step through each draw will be less volatile. Possibly shedding some light on exactly where in the neutral numbers the changes are taking place.

So the plan is to create a version of the script that will index by 1 instead of 7, for the last 7 draws to prove out the concept, which may end up as the refinement for next week's picks, then start looking at the creation of a gui.

Since I am in some need of learning gui coding, it will be a win regardless of what numbers are drawn. After the next 6 weeks go by, if there is no win, it will be back to the sidelines while the above gets completed.

Always good to have a plan

Day pick 3 was a flop, same with day pick 5

Evening pick 3 had 1 number in position, evening pick 5 had 3 numbers in position...

It seemed like a good time to go for the pick 5, as the previous 2 weeks contained no all neutral draws. The evening pick 5 pattern was HCNNN. Since I am essentially playing consecutive games of 1 in 10, I got 3 right. Was that the best shot for the week? Time will tell.

Perhaps an interesting area of study would be the transitioning numbers, the ones with frequencies at the edge of crossing over to the expectancy or crossing from N to H.

With drawing updates, the selection process took less than an hour. The results will take a week to come in. From a time investment standpoint, it is already better than any previous system... and like the others, no hit yet.

This is the only test of the pick 5, but the pick 3 portion has 5 more play attempts... 5 more pick revisions to try. I like the concept of one pick for each game for a week... I should have considered that sooner.

Now I need that hit, calculated or coincidental, to keep interest level high.

Come on Pennsylvania... I have done my part, your turn!

249 years! 7/4/2026 is the semiquincentennial of the good ol' USA!

Picks ready, week 3 of the pick 3 operation.

The day draw I just missed in week 2 was indeed an all neutral draw, I just picked the wrong number from column 2... phase 2 is not very precise... yet.

Might go out on a self funded run at the pick 5 in week 4, it's only $14 extra...

Also considering a self funded run at the Cash 4 Life plan, first week of August... that way the no pick 3 win contingency is covered and I at least get to have a shot. One shot... no pressure...

When studying the results of that neutral near miss, there was no clear indicator that the 8 was a better pick than the 6. End of the day (or week) it is and will always remain a best guess.

This week, the numbers picked were closer to the target. 2 of 3 in position on the mid day and 1 of 3 in position on the evening.

The 1 day stagger per week is the strategy update for the next round. Instead of updating and re running the script tonight, I will do that tomorrow night instead. This will make the 6 weeks remaining take just about 7 weeks to complete.

I am also updating the pick 5 and Cash 4 Life histories in the event they can be rolled out with a hit.

It still comes down to a best single guess, so it might be worth cycling the numbers under consideration through the part of the master data spreadsheet that counts all of the straight and boxed hits, using that result as a tie breaker.

That is tomorrow night's pick strategy...

1. Use the median (2nd quartile) as a range identifier

2. Cycle the combos possible in those ranges to go with the one combo that has the best "record"

On a per column basis, anything learned here DIRECTLY applies to the pick 5 selection process and can serve as a guideline to get a pick (if ever funded by a pick 3 hit) for the cash 4 life.

There is still one draw for both mid and eve to go for this week, and the back test was full of weeks where the NNN draw came in on day 7 of the window.

For this week the plan is set, and for next week I will be taking a look at the standard deviation itself and how it might be applied to the selection process.

This will definitely go down as the longest live trial I have done with any system!

My number for the day game.. 5 6 1

PA's pick for the day game... 5 8 1

2 columns correct...

Not yet there, but an improvement over last week.

In my dabbling in this hobby, I believe we are only given the game set up (pick a combo from 1 in 10, pick 5 of 70 plus a bonus ball of 24) and the draw history. Without said history, why bother?

There are those who believe otherwise, and more power to them. There are those that use the last draw for workouts, there are those that select numbers based on their environment (license plate numbers for example) and those who select numbers based on dreams. There are also players who use quick picks exclusively. There is no real WRONG way to make a pick.

To this end, I continue to study how the numbers were drawn in previous games with the hope that I might be able to gain a better pick for a future draw. I have reached the level in coding and analysis where I spend time disproving my own theories by back testing.

I have run extensive tests on things like mirrors only to see that long term they offered no higher returns than just about any other replacement system (like +111)

Patterns I thought I had discovered in followers did not survive the scrutiny of a back test either.

I have essentially given up on mapping the last draw to the next directly using the numbers.

I still believe that there are answers in the draw histories, even if I am not quite smart enough to capitalize on them.

The system I have currently instead looks at distribution frequencies over a set number of draws, trying to map the last 150 draws to the next 7. 

I do not see another way to make a best guess without using past results as a guide.

If it were easy, everyone would be doing it. 

I rather enjoy the challenge of trying to solve impossible problems. It is taking less time to convert ideas into code, code into spreadsheets, incorporate and validate software.

Even if draws are truly random isolated events, there is some underlying pattern to their distribution.

It is time to up the programming game and convert these latest scripts into a nice looking GUI. And to perhaps take the output from the phase 1 script into a phase 2 script that can be used to make selection easier by adding visualization and to help develop the series of steps necessary to make that best guess.

Although there are as many systems as there are players, I am firmly in the draw history camp.

Happy Coding!

It is interesting how much the distributions change over just 7 draws... 143 of the 150 draws in the model are the same.

The plan is to test the best starting point. I see what the Tuesday night data leading to a Wednesday pick looks like, so the new strategy is to skip the start day weekly...

This week the pick was made on a Tuesday night for Wednesday play. Next week I will wait until Wednesday night to make a guess at a Thursday play... up until I am using Monday night results to make a pick for Tuesday.

So that means this system will get at least 6 more weeks of a live trial.

With that I can compute projected costs...

Already spent $35, will add 6 more weeks at $14 for a total play budget of $119. It only would have been $112, but I played the first week at 1.00 straight and 0.50 box.

So, to make it all back, any 50 cent straight hit will return the play budget along with $131 profit. Plus the box hit at $40 or $80 when a combo is picked with a double.

The system goes by each column, so doubles and even triples are possible. Therefore 3 50 cent box hits @$40 or 2 50 cent box hits @$80 will also return the expense.

This week still has draws through Tuesday evening.

Although any hit, from $40 to $290 will trigger a first shot at the pick5. Also, a straight hit will allow a first shot at the PA Cash4Life.

Hopefully each week will allow for refinement of the selection process.

The plan is also the budgeting process. $14 per week is what I was spending on the Match 6. Sustainable entertainment value. I dropped all play when I started this development process months ago. If there is a hit, I can roll out one or both of the other games for one shot, then go back to the pick 3 only budget.

And, if the next 6 weeks go by winless, it will be an indicator to perhaps cease live play and come up with a better phase 2 selection process. Worst case scenario is that I spent $119 over 8 weeks for some false hope and entertainment.

What was learned so far in the implementation of statistics in Python and the input and output of data was definitely worth the time investment!

Week 1 was $21

This week was $14, subsequent weeks will be $14

I have spent $35 so far without return yet. There are still 6 draws of each remaining.

A straight hit would be profit, a box hit would return what was spent so far + $5

2 out of 3 on the first mid day of this week's attempt, and 0 out of 3 on the evening.

The nice part of the process is that my previous week gets classified so I can see the spread of neutrals that came up. Last week, 3 different NNNs came out, so the number sets produced winners. Just have to figure out which number to pick from each column set a bit better each week.

Going to be coincidence if it works once... gotta remember that...

There is one draw remaining for the first week of testing the new system/concept.

There had been no wins so far. The selection process was not very complex, it was to stick as close to the middle of the middle as possible.

When working on the next attempt, I will begin to employ the different statistics, starting with the median, to help select from the neutral numbers in each column. One step at a time.

A formalized selection process will help with consistency. It is still a process where I am trying to pick 1 in 10, 3 consecutive times. The number of choices is reduced in phase one, but that still leaves roughly 300 possible combos to choose from.

The importance of a definite series of steps is when the pick 5 might be rolled out, it just adds 2 more 1 in 10 guesses, reduced by the elimination of the hots and colds.

At the end of the process it is still a guess, but the best guess I can make.

Every system I work on from here on out will use the phase one classification. Phase 2 will vary week to week until I get it right.

Thus far, had some individual column matches, the best being 2 of 3 on yesterday's mid day. The nice part of playing for a week is that I have a week to think about what to try next, rather than a day.

Going directly for the middle had not yet worked, so now I try using the median as a guide because in a discrete uniform distribution, the median should equal the expectancy, yet sometimes it is slightly lower, and sometimes it is slightly higher.

The new guidelines will be to observe the median (quartile 2) and pick from the group of neutrals as follows...

Median = expectancy, go for the number with a distribution closest to 10%

Median < expectancy, indicating there are more neutrals below the expectancy, so pick from the below 10% group.

Median > expectancy, indicating there are more neutrals above the expectancy, so pick from the above 10% group.

It will always come down to a best guess, and there is still only an 85% chance that the week will have at least one all neutral draw (52% when looking at the pick 5).

The subsequent weeks will be $14 (0.50 straight and 0.50 boxed per draw) so I will probably continue this live test for several weeks. If I go through the rest of June and July with no hits, then it will be time to jump back out and design a completely different phase 2 process. 

The goal is a formal step by step process that might someday be automated, so the program can display not just the classification data, but also generate an algorithmic best guess... because if it gets to that point, it can be back tested!

Tonight will be a draw history update and a run of the script for both mid day and evening, followed by another "best guess"

The distribution percents were not adding up. A number drawn 10 times should have had a percentage matching the expectancy, but instead it was slightly lower.

The issue, using 600 draws in the model, which would have worked if the expectancy was 1/60, but I am using an expectancy of 1/56... easiest fix ever... I just have to change the number of draws in the distribution model from 600 to 560!

Now if a number is drawn exactly 10 times, the percentage matches the expectancy. I eventually catch my own oversights.

While waiting for a pick 3 hit, may as well begin experimenting with a GUI for the output...