Intensive Review Genetic Builder Intensive Review

[Batchboy]

Hi,
How has things been going for you w/ GB Pharoah? Maybe we're not hearing from you because of finding secretive Holy-Grail???
Had couple more weeks of trial usage by leaving my unit on, problem is, the RAM gets stuffed/glutted w/o turning-off unit between jobs, so my smartphone rdp connection to my server became like slow molasses. Is having a button to clear memory and start from scratch w/o having to reboot unit not a good idea for GB? Just wondering aloud.
Did find that picking best results and re-geneticising repeatedly gives better and better results, very nice!
Am not happy about unit having compulsion to make a slew of "empty set" type solution strategies for one exit side, problem goes away if I go for option of wanting a TP. But maybe(?) that was fault of me leaving unit on w/ glutting RAM? Will see when I buy unit and try things anew to see if it's an actual bug or not.
Otherwise, I'm overall pleased that even small memory deployment of GB makes its way. Look forward to custom indicator automatized incorporation as Mark said it is possible and may do, that should be quite enjoyable!
Hope to hear of other folks' use here while I continue saving$ for a unit.
Truly,
Jerry

 

[Pharaoh]

I've been swamped with other projects and the computer I'd dedicated to GB developed issues. The good news is that I've recently freed up another computer (newer and faster!), so when I get a chance to move GB over to it, I expect to get caught up.

 

[John Mendgen]

1.5 weeks into 2 week free trial...

update:

amazing software...
as advertised.

Still unanswered questions, both here and to support, re: handling of missed orders, etc. (though I have been running live on .01 lots trades of one of the 4 free sample ea's, and performing well so far...)

I have my doubts about the validity of the in sample/out of sample theory, which is supposed to minimize the likelihood of optimizing to fit a particular data date range... you generate and test over one part of the data, and then run on part that was held back to see if it is still profitable. It seems to me the ones that are still profitable would be no different from ones that were developed and optimized over both data sets...

However, I tend to think that's a bit unavoidable, no matter what, isn't it?

 

[John Mendgen]

update:

amazing software...
as advertised.

Still unanswered questions, both here and to support, re: handling of missed orders, etc. (though I have been running live on .01 lots trades of one of the 4 free sample ea's, and performing well so far...)

so I did hear back re: above, and though the ea's are not coded specifically for missed orders, as long as the closing condition applies it will try to close with each new tick... so I guess that works.

 

[Mark Fric]

update:
I have my doubts about the validity of the in sample/out of sample theory, which is supposed to minimize the likelihood of optimizing to fit a particular data date range... you generate and test over one part of the data, and then run on part that was held back to see if it is still profitable. It seems to me the ones that are still profitable would be no different from ones that were developed and optimized over both data sets...

However, I tend to think that's a bit unavoidable, no matter what, isn't it?

Hello John,

testing strategies for robustness is one of the most important steps in automatic strategies development. In Sample & Out of Sample test is not the ultimate solution, but it helps filtering bad strategies easily.
Obviously, if strategy doesn't work on unknown (OOS) data, it is most probably curve fitted, and it won't perform.
If the strategy works on OOS data, it is still not a 100% guarantee that it will work also in the future (good OOS performance could be just luck), but the probability is good.
To further test the robustness you can retest your startegies on another timeframe or another symbols.
Next version of Genetic Builder will have also advanced Multi Walk Forward Analysis and Monte Carlo simulation as new tools to test strategy robustness.

 

[Temster]

I've been swamped with other projects and the computer I'd dedicated to GB developed issues. The good news is that I've recently freed up another computer (newer and faster!), so when I get a chance to move GB over to it, I expect to get caught up.

Pharaoh,

Any updates? I am very interested in your testing. I downloaded the trial....but I am thinking 2 weeks will not be enough to evaluate.

Mark,

The ownership option..... is that a one time payment or is there a payment option towards ownership?

 

[Batchboy]

Hi,

Anyhoo, few months ago I couldn't resist to stretch my dollars to buy GB outright, from initial trial you get a coupon for 30% discount.

After running in different ways the best 2 strategies I could find was on 5yrs 5minTF with 100% gain and 18% dd, one did it with MM on, the other did 150%/yr with MM off! The not-so-holy-grail point (and there usually is one or two): low number of trades over that period, the "very boring type EA", lol.

Mark himself had arrived at a superb EA which he tested via myfxbook for.....a year? And he now avails as a product, seemingly secured by dll I think.

My thoughts are that custom indicator pairs are needed to reach near-holy-grail quality on a more regular basis, one that extracts market characteristic, and another that depends on it for an input. Such does exist, for example mqlsoft.com has cycleperiod indicator and a few indicators that depend on it, I have installed those on mt4 so that I can sooner or later use GB's exporterEA to make a file for them. Also, forex-tsd elite indicators are the most advanced, it seems, am occasionally trying those. The downer about custom indicators with GB is that they are not used in "on-the-fly" variable periods manner like the standard indicators in GB. However we can do a few file inputs with some variations of period, of course.....not sure yet if that notion will panout.
So w/o the inside knowledge/feel that Mark of GB has (plus equipment?) it's very tough rowing, very tough journey. I'm using two cores of a friends fast computer too.

Ok, anyone doing any better? Who knows, I might be using it in non-optimum manner? Am trying still to get the feel for what modicum-operandi does best.

Jerry

 

[killerkhan]

nice thread guys, good information.

I got a question about the EA that are generated by the StrategyQuant genetic builder program. are the EA's able to work on multiple currency pairs, meaning do they use magic number and symbol check ? so you can run on more than 1 currency or timefram at a time ?

 

[Batchboy]

nice thread guys, good information.

I got a question about the EA that are generated by the StrategyQuant genetic builder program. are the EA's able to work on multiple currency pairs, meaning do they use magic number and symbol check ? so you can run on more than 1 currency or timefram at a time ?

Hi kk,

They do have magic numbers.

But for each pair the EA usually is developed on one pair, but tester can test on other pair data during its random building if you're looking for strategies that work on more than one pair.
StrategyQuant now has portfolio capability, but I'm not of knowledge about yet as I'm asking Mark about as it seems to me it can't put together other pairs let alone other periodicity strategies, but as I say, am not sure about all this yet, maybe another SQ operator on here can answer.
I just happen to write Mark today about the importance of future portfolio capability versions as my half year experience with SQ shows strategies usually end out being only a couple trades per month for decent characteristics, thus portfolios with at least a few EA's would be behooving to avert public boredom.
Greatest new features is the MonteCarlo robustness tests and the multi-walk-forward-matrix to indicate if a strategy is good for periodically optimizing.
Ok, went on more than I thought I would.
Best,
Jerry B

 

[Batchboy]

nice thread guys, good information.

I got a question about the EA that are generated by the StrategyQuant genetic builder program. are the EA's able to work on multiple currency pairs, meaning do they use magic number and symbol check ? so you can run on more than 1 currency or timefram at a time ?

Hi kk,

They do have magic numbers.

But for each pair the EA usually is developed on one pair, but tester can test on other pair data during its random building if you're looking for strategies that work on more than one pair.
StrategyQuant now has portfolio capability, but I'm not of knowledge about yet as I'm asking Mark about as it seems to me it can't put together other pairs let alone other periodicity strategies, but as I say, am not sure about all this yet, maybe another SQ operator on here can answer.
I just happen to write Mark today about the importance of future portfolio capability versions as my half year experience with SQ shows strategies usually end out being only a couple trades per month for decent characteristics, thus portfolios with at least a few EA's would be behooving to avert public boredom.
Greatest new features is the MonteCarlo robustness tests and the multi-walk-forward-matrix to indicate if a strategy is good for periodically optimizing.
Ok, went on more than I thought I would.
Best,
Jerry B