Optimizing EA
Hello everyone.
Recently I optimized my Expert Advisor(EA) with the Fast Genetic Based Algorithm tool, I set an inicial balance, start date, end date, modeling, symbol, time frame and delays, I chose the parameters I wanted the tester to modify, when it finished, I selected a combination that shown good profit, good profit factor and low drawdown, but when I ran a backless with those parameters, the results shown that this combination actually lost all the balance. To run this test I selected the combination a select the option "Run single test" and keep the setting the same as it were when I ran the optimization. It seems weird to me that in the optimization it show good results but when run in backless it loses the capital.
This Happened because of a bug in the current version of MT5?
Not a bug - just classic overfitting.
The tester did exactly what you asked, not what you hoped.
Worth checking with your EA developer.
Also make sure you "Run single test" in “Every tick based on real ticks” mode for reliable results.
Not a bug - just classic overfitting.
The tester did exactly what you asked, not what you hoped.
Worth checking with your EA developer.
Also make sure you "Run single test" in “Every tick based on real ticks” mode for reliable results.
If I understand correctly Hermes reports that running a single test from an optimization result leads to a different outcome, no changes whatsoever, how can that be an overfitting problem?
Shouldn't run single test from any optimization pass return same exact result from that pass?
Hello everyone.
Recently I optimized my Expert Advisor(EA) with the Fast Genetic Based Algorithm tool, I set an inicial balance, start date, end date, modeling, symbol, time frame and delays, I chose the parameters I wanted the tester to modify, when it finished, I selected a combination that shown good profit, good profit factor and low drawdown, but when I ran a backless with those parameters, the results shown that this combination actually lost all the balance. To run this test I selected the combination a select the option "Run single test" and keep the setting the same as it were when I ran the optimization. It seems weird to me that in the optimization it show good results but when run in backless it loses the capital.
This Happened because of a bug in the current version of MT5?
If I understand correctly Hermes reports that running a single test from an optimization result leads to a different outcome, no changes whatsoever, how can that be an overfitting problem?
Shouldn't run single test from any optimization pass return same exact result from that pass?
You are right.
Though as you already know it's not that simple as most people don't prove technical details to explain and reproduce the issue.
If I understand correctly Hermes reports that running a single test from an optimization result leads to a different outcome, no changes whatsoever, how can that be an overfitting problem?
Shouldn't run single test from any optimization pass return same exact result from that pass?
You are right, if a single test started directly from optimization produces different results (identical conditions) that would NOT be overfitting.
@Hermes Hernández please let us know which modeling mode was used and whether the spread and date range were identical. Some screenshots would also help narrow it down.
Thank you for your interest.
I use same settings for both, optimization and single test, find some screenshots attached. I used open price for modeling because for this strategy I consider that it was not necessary to test every tick, also I selected random delay but I don't think this is the cause of what we see here. I single tested many times and got the same outcome.
Thank you for your interest.
I use same settings for both, optimization and single test, find some screenshots attached. I used open price for modeling because for this strategy I consider that it was not necessary to test every tick, also I selected random delay but I don't think this is the cause of what we see here. I single tested many times and got the same outcome.
Your frustration is understood - still no bug in sight.
"Open prices only" with random delay can easily cause such mismatches (especially when EA logic depends on intra-bar or tick timing).
Try "Every tick based on real ticks" with the same delay - that’s the only way to confirm consistency.
Thank you for your interest.
I use same settings for both, optimization and single test, find some screenshots attached. I used open price for modeling because for this strategy I consider that it was not necessary to test every tick, also I selected random delay but I don't think this is the cause of what we see here. I single tested many times and got the same outcome.
If you do the same test with an EA like Examples/Moving Average.ex5, do you get the same problem ?
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Hello everyone.
Recently I optimized my Expert Advisor(EA) with the Fast Genetic Based Algorithm tool, I set an inicial balance, start date, end date, modeling, symbol, time frame and delays, I chose the parameters I wanted the tester to modify, when it finished, I selected a combination that shown good profit, good profit factor and low drawdown, but when I ran a backless with those parameters, the results shown that this combination actually lost all the balance. To run this test I selected the combination a select the option "Run single test" and keep the setting the same as it were when I ran the optimization. It seems weird to me that in the optimization it show good results but when run in backless it loses the capital.
This Happened because of a bug in the current version of MT5?