Genetic Algorithms, Multi-Dimensional Problem Solving, and Trading

[bwolinsky]

Quote from DeeDeeTwo:

One word = overfitting

"Overfitting generally occurs when a model is excessively complex"

http://en.wikipedia.org/wiki/Overfitting

I disagree that it is complex. It is not overfitting because the model works in real time and has awesome results.

 

[bwolinsky]

Quote from danielc1:

Is this something like Sudoku?

It's a counting problem.

 

[bwolinsky]

Quote from alexvnew:

CPU time is a terrible thing to waste...overfitting....

It's not overfitting because if you had 1000^5 possibilities the only way to solve a 5 dimensional problem like this is to find ranges by narrowing them down with one final exhaustive study, and that is how real financial research in the fields of artifical intelligence is done.

 

[bellman]

When you write 1000^5, do you mean 10^15 (as in a huge number) or is it some sort of lin alg notation meaning a 1000 by 5 matrix? Judging by your phrasing of the question, you mean the latter.

 

[bwolinsky]

Quote from bellman:

When you write 1000^5, do you mean 10^15 (as in a huge number) or is it some sort of lin alg notation meaning a 1000 by 5 matrix? Judging by your phrasing of the question, you mean the latter.

The ranges start at 1 to 1000, so 1000^5 is how many possibilities the genetic algorithms must sort through to find the best ranges.

10^15 is the same value as 1000^5.

 

[bwolinsky]

The correct answer was approximately 124,000.

 

[bellman]

Was there a mathematical method to arrive at this number without knowing a) the data set & b) your genetic algorithms?