Numeric Sequence pattern finder/predictor

[Shingi]

Guys,

Not sure if this is the correct section but I'll give it a shot. I'm looking for numeric sequence prediction program. Say if I input a set of # or #s say three times and see if it can find a pattern on the next #.
say if I input

first input 3 0 0 4 2 9 6
second input 4 1 1 5 3 0 7
third input 5 2 2 6 4 1 8

Now if you're looking at the above # you would notice that I add one to each # every round. I need a program that can find patterns like that and predict the next #. Basically a program that can think and uses a gizillion methods to see if there is a trend/pattern and try to come up with the next set of #s. Is there such program and can I acquire it for use.

Or a second example would be something like this:
First set will be a+b+c = D
second set (D+a)+(D+b)+(D+c) = D
third set will be the same as second set because it just repeats the same process so
a,b,c would be what I would inpute and D would total +a,b,c and so on.

Is there such program? Also if i'm not clear on anything please let me know.

 

[Markbnj]

I'll leave this open in programming because, although it's really a software question, it might spawn some interesting discussion. In terms of existing software and the state-of-the-art, Google for pattern analysis and statistics software. I doubt anything commercial will be inexpensive.

 

[degibson]

The only thing I have to offer is 'look at Neural Networks'. I don't know anything about them myself, but they're the closest thing that I know of to human-like pattern recognition.

 

[Ken g6]

In Shingi's original thread, I mentioned that any patterns found might not be real.

Then I mentioned the possibility that a number of discrete elements, like he gives above, might lead to a finite-state machine. But somehow I expect the elements are more fuzzy, like 3.14 may lead to 4.28 in one case and 3.99 in another?

However, since he seems determined to proceed with this, I'll add a link to Wikipedia's article on AI, a topic with which I'm not at all familiar.

 

[Shingi]

thanks again guys. I'm guessing AI or pattern analysis software is the way to go but I'll take your link up and do research. Anyone have any more ideas please feel free to contribute.

Thanks,

 

[iCyborg]

There are lots of statistical methods for pattern recognition, and some like neural networks are fairly versatile, but the choice of some particular one still depends to a large degree on specific domain knowledge which is not something computers can yet do well. E.g. models used in financial analysis will likely be different than those used for say predicting weather, because a human with a good background in economics can use that knowledge to select a good model. Computers simply don't have that level of understanding and trying "gizzilion" methods won't help either. The best thing to do would probably be to study some methods and decide which ones seem appropriate, there are some programs with stat packages (R, Matlab etc.) to go from there.

Btw, if you want to find a pattern in integer sequences for practical purposes, there's a site with a huge database od various sequences, and you can search it to find those that start with your given numbers. Though it only helps in math when you have some sequence and it looks like it should have a formula, or a name...
http://www.research.att.com/njas/sequences/