In normal regression you try to do

curve fitting by modifying numerical

values of usually a scalar, a vector or

matrix in order to minimize e.g. a sum

of least squares. The result of such an

optimization is a tuple of reals.

In symbolic regression the formula itself

is being rearranged. The result is

a formula along with numerical parameters.

How is it achieved? It can be done with

a genetic algorithm. A formula is a sequence

of objects of class real number, '+', ','

..., 'sin', 'cos', 'exp'. Parentheses

not needed since you're working either

with trees or Polish / Reverse Polish

notation.

The tricky part is that during genetic

operations a so-called well formed

sequence must be generated, but that

is feasible by avoiding steps that violate

this principle. For example, you can use

Polish notation. A binary function results

in one value and expects two values right

of it. Easy to check recursively or with

a stack.

The fitness of such a formula can be

determined by evaluating it at certain

nodes and summing up the squares of

the error. Actually it's the principle

of an interpreter.

Regards,

Joachim

In normal regression you try to do

curve fitting by modifying numerical

values of usually a scalar, a vector or

matrix in order to minimize e.g. a sum

of least squares. The result of such an

optimization is a tuple of reals.

In symbolic regression the formula itself

is being rearranged. The result is

a formula along with numerical parameters.

How is it achieved? It can be done with

a genetic algorithm. A formula is a sequence

of objects of class real number, '+', ','

..., 'sin', 'cos', 'exp'. Parentheses

not needed since you're working either

with trees or Polish / Reverse Polish

notation.

The tricky part is that during genetic

operations a so-called well formed

sequence must be generated, but that

is feasible by avoiding steps that violate

this principle. For example, you can use

Polish notation. A binary function results

in one value and expects two values right

of it. Easy to check recursively or with

a stack.

The fitness of such a formula can be

determined by evaluating it at certain

nodes and summing up the squares of

the error. Actually it's the principle

of an interpreter.

Regards,

Joachim

I would look at the latest paper by Nuno Lourenco.

Dear Pimiskern,

Thank you very much for detailed description on symbolic regression. I would be glad if you please share a Matlab code in which symbolic regression is implemented. Form theory of symbolic regression, it is very difficult for me to understand it. 

Thanks

with

Warm regards

Dear Ryser-Welch,

It is difficult for me to identify Prof. Nuno Lourenco. I request please share any one research paper for the same.

Thanks

with

Warm regards

Unfortunately I have no code in Matlab.

Here is a paper by Nuno Lorenceo et al.

Studying the Properties of Structured Grammatical Evolution

ECOS-CISUC Technical Report TR2015/01

Nuno Lourenco, Francisco B. Pereira, Ernesto Costa

https://www.cisuc.uc.pt/ckfinder/userfiles/files/TR%202015-01.pdf

An introduction:

http://creativemachines.cornell.edu/sites/default/files/Science09_Schmidt.pdf

Regards,

Joachim

Hello Ghanshyam Tejani

 The following software are available for implementing symbolic regression:

GPTIPS software:  is a free, open source, MATLAB based symbolic regression     software

Eureqa  software:  implements symbolic regression

DataMelt  software:  implements symbolic regression in Java and Python

RGP software: implements symbolic regression in R

You don't need to write your own code to use these software.

Eureka is freely downloadable from (M. Schmidt Eureqa web page) http://creativemachines.cornell.edu/eureqa