Tip: Avoiding lopsided optimizations
-
When running a genetic optimization, sometimes the best results converge towards an inputs set that produces a very small number of very successful trades. This is usually a bad thing, because few very successful trades is statistically less reliable than a lot of semi-successful trades.
To avoid this optimization problem, you can code a custom criterion that takes trade number into account, but if you are lazy like me, you can use a "handicap": create some extra blocks that make 100% bad trades -> now the optimization discards all results with too few trades, because those trades cannot offset the intentionally bad trades. Input sets that produce a lot of good trades will have the upper hand in optimization.
Example of a simple handicap model:
https://fxdreema.com/shared/Jdk6zDAtb -
@roar Very good contribution! I'm going to put it into practice, thanks!
-
Interesting, as always. Thank you
-
very good! Thanks Roar.
Thatll go well right along with a few of the EA termination blocks I use such as after X failed trades, terminate EA....or after X drawdown, terminate EA....not enough profit over time...TERMINATE!
