by Jerzy » Tue May 13, 2008 2:41 pm
Joe,
I believe that there could be several ways of implementing a cluster pattern filter. Midas has an idea of a cluster pattern filter which would be integrated with a predictor. Probably he also knows how to do it. Right now I am proofreading the Excel file he posted on the Forum. It contains a list of all possible variations of cluster patterns The patterns contain 0's so their list is long. I will finish proofreading probably tomorrow.
His idea of the cluster pattern filter integrated with the predictor is very attractive but would be probably more difficult to implement than a basic cluster filter. And I feel that simple cluster filter would be also easier to use. So it would be great if two cluster filters could be implemented, one basic and one advanced filter.
I believe that number of rows and columns in the panel must be used as parameters in calculating the clusters (it seems to be unavoidable). Complex calculation is needed to identify clusters of 2 and 3 as well as single separated numbers. The clusters and separated numbers are surrounded by isolating numbers which need to be included in the calculation. All numbers in the cluster are grouped around one central number (a pivotal number).And each of the number in the panel will play a role of central number for all possible cluster ( probably with the exception of last several numbers, eg. number 45 in 6/45 game cannot be a central number for any cluster, number 44 will be the central number for the last cluster of two, and number 43 will be the central number for the last cluster of 3).
A separated number can occupy 3 rows and 3 columns ( 1 row and 1 column shared with the neighboring cluster).
A cluster of 2 or 3 will occupy more rows and columns. Mark some clusters on the game panel and see how much space they occupy and how this space could be defined in the most simple way.
I know my comments are rather general, probably you already have more specific ideas.
Go ahead with this filter, Joe.
Regards
Jerzy