My first GAT test

[peteros]

This is my first "real life" test of the GAT-engine.

I compiled my notes on this page http://www.eprdata.com/gat-test.html hope it's not to confusing.
All amount is in local currency.

[lottoarchitect]

Hi peteros, I love the way you present your analysis! One question, did you use the initial prediction of 16 numbers for all the 20 test draws? If so, this is wrong to do because GAT produces dynamic predictions for each new additional draw. However it is apparent the temporal effectiveness of that GAT table which gives many 5+ hits over the next 9 future draws but it then loses quickly the efficiency. Still the correct way to make such an analysis is to make a new run using the run factor for each of those 20 tested draws. To make this clearer, you initially run GAT to predict draw W1-1. You have run factor 0 and you decided obviously to use GAT #72.428.957. You get the prediction of that GAT. Now, in order to compute draw W1-2, you add the draw W1-1 to the history, set run factor to 1 and run GAT again. The new prediction you get from the GAT 72.428.957 is what to use, not the initial set of 16 numbers predicted for W1-1. The same goes on for all the following draws. So, for S1-1, you now have run factor 2 and run GAT and pick the new numbers of GAT #72.428.957. From your analysis I conclude that you just used the initial prediction performed for W1-1 draw in all the other draws. This can be effective only for a few draws ahead (the temporal effectiveness) but not beyond that.
Apart from that, playing a wheel of 16 numbers in only 12 lines each is not really productive because as you can see, most of the time we hit only up to 4 numbers even if the prediction contains 5 or 6 numbers. I'd suggest to opt for a lower amount of total numbers played in order to play cost budget wheels. Can you do the same calculations with bigger wheels (as many lines as needed to properly cover the m condition) to check the overall cost vs winnings?

cheers
lottoarchitect

[lottoarchitect]

I also had a look at the Swedish lotto game. It looks like a 7/35 + 4BB (4 bonus balls?!). Have you included bonus balls too in the analysis? Is there also a download link of history draws for that game? I can't find it at the official website.

[pusha]

Hi peteros,

nice result, the swedish lotto seems to give low winnings for 3/4 and 5? Nevertheless, as we can see GAT can predict pretty good. I'm testing longterm GAT too, i wonder which cycle of draws and size of draws is the right to go. Till the testing period is finished, keep it up .


cheers,

pusha.

[peteros]

To: lottoarchitect
Yes I did use the first prediction for the entire 5-week period because I want to find a good strategy. If a GAT regularly gives good hits over more than 1 draw a good strategy might be to play the same numbers for 2,5 or 10 weeks. One strategy might to play 10-lines for 5-weeks AND play a new GAT in the same way for every draw or every week instead of playing 50,60 lines for one time only, then there might ( or not ) be a bigger chance to win.

As you could see there were some sixes and some fives in the first half, after that there was very few hits in the last draws. However all of the tested wheels missed the sixes
At the moment I'll just try to find a good strategy for the game.

The GAT I used was the only one that had 100% 4 or better in the lowest amount of numbers(16) there was a few 95% GATs with 15-numbers however none was close to a good win.

”Now, in order to compute draw W1-2, you add the draw W1-1 to the history, set run factor to 1 and run GAT again.”
I can't add do this, because of the 2-draws is direct after each other, so I have to use a run factor of 2 and input both draws from that day.

The 12 lines used is 1. for my budget. ( Try to find a few more people to play with )2. One coupon for that lotto is 12 tickets. I have also checked with 7 numbers to play as single tickets, but for this run there was no success, but I'll try the same strategies for at least 10 draws to see if I can find a good strategy.

Yes, the Swedish lotto is 7 out of 35 numbers with 4 bonus numbers, no bonus was used in the calculations but is included in the database I use in GAT engine. I don't want to find alot of 4-hits where there is bonus ball alone that hitted. If I use the bonus in calculations I have to go for a minimum of 8 hits to be sure at least 4 numbers is regular numbers.

You can find the draws in the official website at www.svenskaspel.se under menu ”resultat” - Lotto Lördag (Saturday) or -Lotto Onsdag (Wednesday), however you have to choose a specific date and can't just download a list of draws, that's why I only have year 2011 in my data

[peteros]

To: pusha

The Swedish Lotto pays at 4,5,6,6+b and 7 numbers. At 4 correct it pays about (usually) 3-4 times the cost of 1 ticket for 2 draws (2 draws 2times/week). 5 correct pays about 5 times more, 6 correct 25 times more than 5 hits and 6+b additionally 10-20 times more than 6 hits. 7 correct is a million or two .

The official payout table:

7 35%
6+1 12%
6 7%
5 12%
4 34%

How much the payout is for each winning ticket depends of the amount of winning tickets at the same level.

[lottoarchitect]

Hi peteros,

As you could see there were some sixes and some fives in the first half, after that there was very few hits in the last draws. However all of the tested wheels missed the sixes

That's the temporal effectiveness of the prediction (there is another thread discussing this). Quite often a prediction performed for a given draw can hit well up to a few draws ahead, especially if it don't hit well at the draw it was aimed for. After that short period, it degrades. It is a viable approach although not officially suggested; it needs more investigation to properly recommend this approach for use, but already a few people suggested that it can give good hits. My opinion is that it is best overall to run the engine with the run factor (making a new prediction for every new draw).

I can't add do this, because of the 2-draws is direct after each other, so I have to use a run factor of 2 and input both draws from that day.

Indeed, for that reason you can rely on the temporal effectiveness of the prediction for the Wx-2 or Sx-2 draws.

About the wheels, surely having 4+ correct numbers almost all the time generally will need more numbers to be picked. However the cost vs wins presented rely on blocks that don't really form a wheel (4if5 needs 28 blocks - my record!, 4if4 needs 76 and 3if3 doesn't really offer an edge when having more correct numbers - not optimized for such a task). The bill can't really fit in 12 lines and 16 numbers! This is why I query the possibility to virtually run those numbers in proper wheels for the size and m condition (use a 28 block wheel for 4if5 etc). If you have the time, it would be interesting to see the cost vs win ratio with these wheels and the same hit results achieved by the engine. I'm almost certain that the increased cost of playing these wheels will be amended by the overall earnings produced.

cheers
lottoarchitect

[pusha]

Hi peteros,

am i right that you have made the run for the hole 35 numbers and scaled down to 16 numbers and than looked for a GAT that gives 100% for a 4?

I've been testing long term GATs with different results, overall it's hard to find a GAT with frequently hot cycle. It really depends on your playing period. For my test i've been using a period of a year where each weeks have 2 draws. And therefor the amount of hot cycle's are really high.

Maybe you want to test with 4if4, 5if5 or 6if6 with some filters like SUM, Odd/Even and Position of the most concentrated expected value of your lottery in WG.

Cheers,

pusha.