Optimization flaw

New comment
 

I'm running an Optimization for an EA I built and even though I have "Skip useless results" UNCHECKED the strategy runner is not running every possible scenario (or it is and is just not showing me). The strategy runner should be running 900 different combinations but it's only doing 100 passes.

Has anyone encountered this or heard of anything like this?

Thanks,

Pat

 

 "Skip useless results" UNCHECKED 

There is such a checkbox in Tester?

 
blogzr3 wrote >>

There is such a checkbox in Tester?

Yes, Under the Optimization Results, if you right click on one of the results, it is an option

 
FXpipclash:

I'm running an Optimization for an EA I built and even though I have "Skip useless results" UNCHECKED the strategy runner is not running every possible scenario (or it is and is just not showing me). The strategy runner should be running 900 different combinations but it's only doing 100 passes.

Has anyone encountered this or heard of anything like this?

Thanks,

Pat

Hi Pat

Untick the genetic algorithm box on first tab page it does something fancy and if you are getting wins it moves to other areas of parameters. I think you are supposed to look at the results then run without genetic algorithm to home in on the best parameter set.

 
Ruptor wrote >>

Hi Pat

Untick the genetic algorithm box on first tab page it does something fancy and if you are getting wins it moves to other areas of parameters. I think you are supposed to look at the results then run without genetic algorithm to home in on the best parameter set.

Ruptor,

Much thanks for your help, this has boggeled my mind for some time now.

 
Ruptor:

Hi Pat

Untick the genetic algorithm box on first tab page it does something fancy and if you are getting wins it moves to other areas of parameters. I think you are supposed to look at the results then run without genetic algorithm to home in on the best parameter set.


Is your statement correct? Look at this: https://www.mql5.com/en/articles/1409

 
arachnode.net wrote >>

Is your statement correct? Look at this: https://www.mql5.com/en/articles/1409

I'm going through this article as we speak. To help my comfort level, I like to see all of the results and what works and what doesn't work. I will begin to play with the genetic algorithm function and am sure over time figure out the most efficient way to use it. If you have any personal recommendations or suggestions I would appreciate it.

Thanks for all of your help,

Pat

 
I think it makes sense to iterate through all possibilities and analyze, but after you use the GA optimizer to find out if you're on the right track or not. I had wondered about the differences in results, and the article suggests that they are more or less the same.
 

Hi Guys

My experience with the back tester in general has not been good. It is a good idea but the implementation is rubbish. I don't bother doing long runs because it only takes the data to be out and the time is wasted. I have optimised using the genetic algorithm and had no profitable runs and yet it made a profit on a single run not using the optimiser. I mainly use the back tester to check the code works and to set TP & SL. I usually know roughly what area the strategy parameters should be in since the EA is based on visual observations on indicators to start with but you have to find your own way in the end just like the trading itself.

Was the genetic switch the answer to your missing trades?

 
Ruptor wrote >>

Hi Guys

My experience with the back tester in general has not been good. It is a good idea but the implementation is rubbish. I don't bother doing long runs because it only takes the data to be out and the time is wasted. I have optimised using the genetic algorithm and had no profitable runs and yet it made a profit on a single run not using the optimiser. I mainly use the back tester to check the code works and to set TP & SL. I usually know roughly what area the strategy parameters should be in since the EA is based on visual observations on indicators to start with but you have to find your own way in the end just like the trading itself.

Was the genetic switch the answer to your missing trades?

Yes the genetic switch was the answer to my missing trades, however, I still have issues with the strategy tester. Back tests of the same time frame I conducted last week are returning different values this week (for the same time period of course). There seems to be something, "off" with the whole strategy tester sometimes. Also, after running optimizations, I will input the parameters from an optimization into a backtester and it will show different results than what the optimization showed. It's rather odd...

 
FXpipclash:

Yes the genetic switch was the answer to my missing trades, however, I still have issues with the strategy tester. Back tests of the same time frame I conducted last week are returning different values this week (for the same time period of course). There seems to be something, "off" with the whole strategy tester sometimes. Also, after running optimizations, I will input the parameters from an optimization into a backtester and it will show different results than what the optimization showed. It's rather odd...

Hi Pat

Sounds like you are back testing on a demo account, don't bother stick to live accounts the data has to be more reliable since money is involved. To get consistency people set up an MT4 that is off line and can't connect to the broker. They download the history files and after that keep it off the net so it can't update the history and get screwed up. The different results you are getting could be the results of different data being downloaded with every test. Take your pick of the methods but in the end the only way to be sure is forward testing isn't it?

12
New comment