Flabber
First, I am not your mate.. whatever that implies. Second, your feelings seem to have been bruised
so may I suggest a little corner time out. Markov chains, ie followers as you call them are very simple
to generate for almost any data type. The steps are a base-10 to 3-digit code. Each digit 0 to 9 is
coded into a specific code, example 212=5. The program first converts the actual line, ie "set of
numbers" into it's lexicographical index. The index is zero padded so that it's always the same length
depending on the matrix, a 5-39 =6 digits in length. 000001 to 575757.
Each of the lex-I digits is then converted into it's 3-digit code called steps. The first digit in the step
has a range of 1 to 3. The second and third both have a range of 1 to 2.
Now, provided your able to understand the above you should come to the conclusion that I end
up with 6 steps, each with 3 digits for a 5-39 game matrix.
day date number set lex-I Steps
SAT 06/02/18 05 17 19 34 38 289035 122|321|312|121|211|212
FRI 06/01/18 03 14 16 21 36 184188 111|321|221|111|321|321
THU 05/31/18 04 05 21 22 28 203786 122|121|211|311|321|222
WED 05/30/18 21 28 30 34 39 566765 212|222|222|311|222|212
TUE 05/29/18 11 13 21 24 29 461596 221|222|111|212|312|222
Now maybe you can see where the Markov chains come in. There are 18 vertical columns where
the first digit in each step has a range of 1 to 3, whereas the 2nd and 3rd are limited to 1 and 2.
Each vertical column is loaded into a string beginning at a random starting point. The reason
for this is so that every line generated uses slightly different data and to exclude the program
from including the set I am trying to predict from being used as part of the analysis.
I am experimenting with using a fixed starting value for the strings, say 10 games back and then
randomly generating the value of data points added to each string before the Markov calculations.
After the data is processed comes the tricky part, making the best prediction using the results gotten
from the Markov chains, not as easy as one might think.
Once all 18 columns have been processed and the values selected it's a reverse process from prediction
back to steps, steps to base-10, base-10 to Lex-I and then Lex-I to the actual line/set. Takes around 2
seconds to generate the 100 lines/sets of 5 numbers.
Sorry, no BS here, just my simple spin on basic Markov chains that seems to be working far better than I
ever expected. My state uses PRNG to draw numbers which may or may not have something to do with
the output.
In testing ball dropper games MM and BP it has not produced a JP but little testing was given to those games.
Who knows, there is still the chance that a mistake in the code exist that would spoil the whole thing.
I've been using these types of data conversions for at least 15 to 20 years. Several years back I released
a free software called T-Lex which has everything I use here except the Markov predictor.
RL