01-30-2015, 05:03 PM
Has anyone who visits this site done some work or studied Genetic programming? I am curious to see programming example on how GP works to, for example, develop a good regression model.
Namir
Namir
01-30-2015, 05:03 PM
01-31-2015, 01:12 PM
01-31-2015, 03:09 PM
That is a very interesting paper - and topic.
01-31-2015, 09:26 PM
The trick is to be able to dynamically alter the expression tree! It's NOT easy.
The alternative approach that I have used is to apply test regression models between multiple variables with EACH variable having a set of different possible transformations. this approach can test hundreds if not THOUSANDS of regression models. The approach is deterministic as it gives you the same exact results whne repeated.
Genetic Programming (GP for short), variant of Genetic Algorithm, has a stochastic element in it. So repeating calculations, which involve random numbers, most likely give you different results.
Matrices can easily model graphs and trees--both of which contain nodes. The graphs in GP contain nodes that are either constants, variables, operators, and even functions. Dynamically knowing what sub-graph to add and what part of the current graph to remove remains a mystery to me. My goal is to learn and master teh algorithm that dynamically grows and prunes expression trees.
Namir
The alternative approach that I have used is to apply test regression models between multiple variables with EACH variable having a set of different possible transformations. this approach can test hundreds if not THOUSANDS of regression models. The approach is deterministic as it gives you the same exact results whne repeated.
Genetic Programming (GP for short), variant of Genetic Algorithm, has a stochastic element in it. So repeating calculations, which involve random numbers, most likely give you different results.
Matrices can easily model graphs and trees--both of which contain nodes. The graphs in GP contain nodes that are either constants, variables, operators, and even functions. Dynamically knowing what sub-graph to add and what part of the current graph to remove remains a mystery to me. My goal is to learn and master teh algorithm that dynamically grows and prunes expression trees.
Namir
02-04-2015, 12:42 PM
Namir,
while not running on an HP calculator I have used a program which was intially called "Eureqa" (like the old Borland math tool) and now sails under the name "Nutonian formulize"
(see http://www.nutonian.com/).
This is a handy tool to create analytical approximations of data sets. It constructs and evaluates equations from basic building blocks using a genetic algorthm. You can watch it at work and finally select a compromise between accuracy and complexity for your approximation equation. Maybe you can get some ideas for an implementation on a calculator. I would expect rather high memory requirements though.
Martin
while not running on an HP calculator I have used a program which was intially called "Eureqa" (like the old Borland math tool) and now sails under the name "Nutonian formulize"
(see http://www.nutonian.com/).
This is a handy tool to create analytical approximations of data sets. It constructs and evaluates equations from basic building blocks using a genetic algorthm. You can watch it at work and finally select a compromise between accuracy and complexity for your approximation equation. Maybe you can get some ideas for an implementation on a calculator. I would expect rather high memory requirements though.
Martin
02-04-2015, 02:44 PM
(02-04-2015 12:42 PM)Martin Hepperle Wrote: [ -> ]while not running on an HP calculator I have used a program which was intially called "Eureqa" (like the old Borland math tool) and now sails under the name "Nutonian formulize"
Borland's Eureka -> Real Software's Mercury
:)