colleagues decided to highlight the issue in a separate topic. When optimizing parameters there are a number of standard features, as well as “Max user” criterion that allows the selection relative to its target, even if only in the direction of the maximum. unfortunately the result of optimization is the maximum growth of the target function during the training period and as a result there are such parameters for the adviser that are relevant here and now but not for the future of its work. after all, the parameter you are looking for is already maximum. if we take the balance curve as a standard, then in fact we get such parameters of the adviser at which he has already scored everything that he could and his next work usually leads to a decrease in the curve and losses if we assume that the balance curve graph has a wavy appearance. then it dials and merges. so i am deeply convinced that you need to look not for the maximum point of balance, but you need to look for the point of the beginning of the climb by means. in fact, the balance curve should look something like this.:
i duplicate a post from another branch. So, when optimizing, we need to find the starting point for the growth of the balance sheet. The practical application of this looks something like this:
This picture shows the work of a simple counter-trend indicator, but an elementary adviser on mashki will also get its relevance, because no matter how you look at it, all the advisers periodically gain then merge. finding the points of the beginning of the growth of the balance curve will be more guaranteed that this growth will continue than when it is already at the highest height. actually, the question is, what is necessary to maximize the balance calculation function so that the balance curve itself has the above form????
Just a couple of weeks ago, I managed to make a complex evaluation function of the model during training that includes three separate estimates and I will say that I am extremely happy with the result, so here I think that you need to think about the function when maximizing which the balance argument would have a similar form. what do you say, gentlemen???
blaze
Make a takeprofit equal to the stop loss and it will always mirror.I dug in this direction, it’s not much use, just in the same periods, then it works in the future after such optimization, then it doesn’t work.
rrr5
and what criteria does it take into account?
nikelodeon
You could not specify in more detail what the criterion is, I do not observe it yet :- ( And how to work with it in a nutshell literally….
renat
Try the standard complex criterion that appeared in the latest beta versions.
we specifically developed it for a more conscious choice instead of optimizing for single characteristics.
greshnik1
We optimize to the limit, “flip” and look at the results on new data :)))Post what happened
nikelodeon
As the teachers of the “Forest of Hands” used to say, I don’t even know who to ask.
Alternatively, build a reference curve in relative units. Calculate the difference between the standard and the balance curve and try to minimize it. Who will say what???