READ-ONLY

The forum is now a read-only archive.

For bug reports & platform questions → [email protected]

Our community lives on Discord and YouTube — come join us!

why using random generation?

3 replies

ybhx0315

Customer, bbp_participant, community, 27 replies.

Visit profile

10 years ago #115279

hi all,

 

I am just having no idea of this question which confuses me from the first day I used SQ.

random generation merely generates strategies randomly without a direction or improvement. Genetic generation tend to optimize your strategies during build.

 

Then the question is, why do we bother to use random generation at all? Any advantage does it have over genetic?

I saw that autotrading academy is using random. Is it just for demonstration? or for other reasons?

thanks

0

Tomas Matejka

Administrator, sq-ultimate, 2 replies.

Visit profile

10 years ago #137928

Genetic programming tends to be biased towards local optimum. You are optimizing using fitness function on generated population. That’s where random generation helps because it adds more randomness and can be sometimes better finding nice setups under set conditions (building blocks, money management). To me it’s about finding a good strategy using random generation and then try to improve it. It’s only personal preference what to use

0

Rom

Customer, bbp_participant, community, 29 replies.

Visit profile

10 years ago #137934

Ok, but genetic algo should try to converge to built objectives.  It doesn’t.  When the number of built objectives is >2 the the program fails to create the seed.  From what I understand, the seed should be random and the algo supposed to converge to objectives.  If I understand the objectives, i.e., I know what kind of properties I expect from the built, the program should be able to help me.  Now, because of the fact that the initial generation is random, the “evolution” may fail, and that’s OK, I may try it again.  Unfortunately it doesn’t happen that way.  

The alternative ( random generation with constraints) is doomed to failure from mathematical point of view.  

0

ybhx0315

Customer, bbp_participant, community, 27 replies.

Visit profile

10 years ago #137937

Genetic programming tends to be biased towards local optimum. You are optimizing using fitness function on generated population. That’s where random generation helps because it adds more randomness and can be sometimes better finding nice setups under set conditions (building blocks, money management). To me it’s about finding a good strategy using random generation and then try to improve it. It’s only personal preference what to use

If genetic evolution tend to find the local maximum, fine, the whole evolution will restart after n generation of stagnation. It still keeps the good aspects of randomness and I still have no reason to give any preference to random generation method.

0

Viewing 3 replies - 1 through 3 (of 3 total)