by Jerzy » Mon Sep 24, 2007 4:37 pm
Joe,
The next filter idea that is worth implementing is a Cluster Filter.
When you look at the marked panels on the playslip, you can see that on some panels the ticket numbers are separated (or isolated) from each other, on other panels they are clustered into groups (or a group) of 2, 3, 4, 5, or maybe 6 numbers, and on certain panels there could be a combination of separated numbers and clusters. Numbers in a small cluster are adjacent to each other vertically, horizontally or diagonally, but in a large cluster some of them are connected with others through the adjacent numbers.
We can calculate a cluster number, which is the count of isolated numbers plus the count of groups of numbers connected in clusters.
The cluster number range can be from 1 to 6.
Or, we can find cluster pattern, a composition of clusters (noticing their sizes) and separated numbers. The second option makes more sense if you want to build a good filter. The cluster number can be the same for certain patterns which have different occurrences and some of them could not be safely removed but some of them could be. For example: the cluster patterns 2-2-2, 3-2-1, and 4-1-1 have the cluster number 3, but they differ considerably in their occurrrences in the history data and in the full package of tickets.
The cluster number and cluster patterns depend very much on the total pool of numbers in the game and on the shape of the game board.
Expert Lotto is already configured for total count of numbers in the game, count of numbers drawn at each drawing, and the shape of the game board. It uses this configuration when working with existing filters, like Panel Filter or Number Filter (the shape of the game board is clearly displayed at certain filters and you can select numbers on it). So probably implementation of Cluster Pattern Filter in Expert Lotto is possible and might be even easy.
The following cluster patterns are possible:
1-1-1-1-1-1
2-1-1-1-1
2-2-1-1
2-2
3-1-1-1
3-2-1
3-3
4-1-1
4-2
5-1
6
I used Excel to do statistical analysis of cluster patterns in the history of my game.
I placed past winning numbers in the columns B, C, D, E, F, G, H (in column A the date) and in the next 45 columns I counted the occurrence of each number in the combination if it appeared of the game board as a separate number. Conceptualization of the formulas for counting the occurrences of separate numbers and clusters of 2 numbers took me some time.
Next, I identified all clusters of 2 numbers that are possible, as they appear on the panel. To be sure, I just marked them with a pencil on the play slips. After that I entered the pairs of numbers which form clusters as the headers of colums (pairs from 1-2 to 44-45). I calculated the occurrences of these pairs in a given combination if they appeared on the game board as a cluster.
I have not calculated directly the occurrences of clusters larger than 2 numbers because Excel has not enough columns for that.
I calculated total count of separate numbers and total count of clusters of 2 numbers in a combination, and arrived at the correct cluster pattern in a combination by a simple process of deduction.
5-1 if there are no clusters of 2 nrs and there is 1 separate number.
4-2 if there is only one cluster of 2 nrs and no separate number.
4-1-1 if there are no clusters of 2 nrs and there are 2 separate numbers.
3-3 & 6 if there are no clusters of 2 and no isolated numbers. Unfortunately, because I could not count the occurences of clusters of 3 directly, I had to consider both groups jointly. As both groups have very low occurrences and both are good candidates to be filtered out, it does not really matter. But if you do paterns of 3-3 and 6 separately it will be an elegant solution.
3-2-1 if 1 cluster of 2 nrs and 1 separate number present.
3-1-1-1 if no clusters of 2 nrs but 3 separate numbers present.
2-2-2 if 3 clusters of 2 present.
2-2-1-1 if 2 clusters of 2 and 2 separate numbers.
2-1-1-1-1 if 1 cluster of 2 and 4 separate numbers.
1-1-1-1-1-1 if all numbers are separate.
To help you understand this better, I attach the copy of the Excel sheet that I used for calculation of cluster pattern statistics. If I replaced the history data with combinations from the package, I could delete the combinations with cluster paterns which have low occurrence. But this method of filtering is not good enough. I hope you will create something better.
If you need any additional information, let me know.
Jerzy
You do not have the required permissions to view the files attached to this post.