Praise for MC incremental genetic algorithm

Questions about MultiCharts and user contributed studies.
glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Praise for MC incremental genetic algorithm

Postby glam_100 » 20 Jul 2007

The incremental genetic algorithm in MC is the best I've ever seen implemented on commercial GA products.

The quality of the solution is so good that it is a truely viable alternative to brute force. The basic genetic algorithms used by other GA products are pretty much unusable in my opinion and the often gave the wrong solution.

Now if we can specify the fitness function in EasyLanguage, it would make a very powerful GA optimizer

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 23 Jul 2007

Thanks glam_100,
I agree that the GA algorithm we use is one of the best.

Regarding fitness function. Could you explain how you intend to use it?

Maybe it is easier for you and us to create several criterions for search and you would be able to indicate weights of these factors? For instance you need max net profit and highest %winners.

Please describe.

Guest

Postby Guest » 23 Jul 2007

good ideas
- why not let the user describe his GA objective with a PowerLanguage function? One could even write combinations of min/max functions that would satistfy nearly every criterion.

all the best!

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 23 Jul 2007

Hi Andrew,

I suggest the TS support team can do this:


1) Create a set of out-of-the-box fitness criteria. You already have that done and you may want to throw in a few more exotic ones derived from the standard criteria such as NetProfit/MaxIntraday DD, Expectancy, NetProfit * Winners/MaxIntraday DD ...etc.

2) You then create a table that let people to add the fitness criteria of interest from 1) (more than one criteria can be chosen), and allow a percentage weighting scheme to specify the importance of each criteria. For example if a user has 5 criteria added to the fitness table and each are of equal importance, then the weighting would be 20% each for the 5 criteria.

More advanced feature:

3) Allow the user to create custom fitness criteria using EasyLanguage and add to the default fitness criteria in 1)

You see all existing 3rd party TS GA tools such as TSGO, Optimax, and Grail all convert the original strategy into EL wrapper code, and then the users would run the converted wrapper code in TS to do their GA opt. It is tedious but a side benefit is the user can always tinker fitness criteria in the wrapper code.

Although the integration of GA into MC is a huge time saver and removes the need to do code conversion there is a community of advanced TS GA users who are using their own custom fitness criteria with the existing 3rd party GA products. With the current version of MC they can't do that.

In my opinion the highest priority is 1) and 2) because standard single search criteria such as net profit or profit factor alone is completely useless in GA. At least we will need to be able to specify a *List* of criteria such as net profit *and* profit factor *and* number of trades *and* return on account ...etc

Then having custom fitness criteria is nice to have...

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 23 Jul 2007

Hey am I asking for too much?

It may be a lot of work but these are real issues faced by many TS users.

I really wish TS support to succeed. TS has changed itself from a research tool developer into a brokerage and all they want is trading commision. They've lost all interest to improve the research capability of their product and the users can totally feel it! You see improvement suggestions sitting there for YEARS even though they're relatively easy to implement (say range bar chart). I think they have not materially improve the research capability of their product since 2000i except for the introduction of Look-Inside-bar and intra-bar-order. You can see that clearly when a lot of people are still using 2000i and have no incentive to move to TS8. So TS is more like 7 or 8 years outdated (or we should say they were way ahead back then to their credit).

TS is a great product. The only issue is the attitude. I think there is a real pressing need for companies like TS support to pick up what TS has left off. TS could have throw in the 3 big improvements (multi-core, GA opt, and portfolio test w/ money management) quickly given their huge development resources but they just won't do it. It is companies like TS support that are giving us the last hope! No exaggeration!

I think MC will already surpass TS for research purchase once you fix the speed issue and roll out Look-inside-bar (and of course fixing all the bugs). The only reason now I still stick with TS is the speed and LIB. Then when you improve your GA with fitness criteria list and custom fitness function, you will make the TS + 3rd party GA combo irrelevant given a 3rd party GA tool alone cost hundreds. Then finally when portfolio testing w/ money management comes out TS will be left in dust. There is simply no reason to stick with TS at all...

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 27 Jul 2007

glam_100,
Thanks for the suggestions. We will analyze it carefully.


Return to “MultiCharts”