Has anyone made Automatic Virtual Self-Optimization for their robot? - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
You need to understand what you're trying to explain.
Why don't we all chip in and give you a teddy bear?
Why don't we all chip in and give you a teddy bear?
Can't you just write something on the subject for fun for once?
,..
Is it worth applying it ?
Once you start doing it, you realise that you can't reach the functionality of a tester - there are no ticks (not in the sense of no ticks at all, but in the sense that you'll get tired of modelling them). Therefore the solution is incomplete. Besides, each modification of the strategy will require a serious modification of the embedded optimizer. So we get a solution that is not all-purpose. That is why an idea arises - why not use the second terminal for automated optimization? And when you think carefully how to do it, you forget about it. What is the problemof launching manualoptimization once a week?
How to explain it... with a monotonically changing pattern, self-optimisation works. For example, if a straight line under a slope grows and for the TS you just need to update the data (parameters), recalculate for new values, and all that sort of thing
in the market it's a guessing game in any combination, because the patterns change by leaps and bounds and in a dramatic way
and if you've found a period for self-optimization, it means you've found cycles to correct the parameters and you don't need a self-optimizer anymore
self-optimisation is the equivalent of a moving average.
This is the key factor why I gave up on it about 3 years ago. Optimised for history, from Monday the market became different - newsy, and the system is simply not ready for it. From the next Monday the situation is the opposite and the settings are not suitable again.
The answer fromfxsaber gave a thought: "I need to earn more on the quiet part than to lose on the jump.
Probably need to rethink the entry logic itself, which I will do closer to the snows. Never knew it was called machine learning :)
I am saying that self-optimisation is not necessary at the trading stage.
if a boxed version, why not? just so the user doesn't get bogged down with optimisation.
If it's a boxed version, why not? just to keep the user from getting bogged down with optimization.
yes please )) I'm just talking about the method itself in its bare form, which will not greatly improve TC if there are no regularities as such
i.e. purely conceptually
This is the key factor as to why I gave up on it 3 years ago. Optimised for the story, from Monday the market became different - newsy and the system is simply not ready for it. From the next Monday the situation is the opposite and the settings are not suitable again.
The answer fromfxsaber gave me something to think about: " I should earn more on the quiet sector than to lose on the jump.
Probably need to rethink the entry logic itself, which I will do closer to the snows. I never knew it was called machine learning :)
Especially yes, when the sliding optimization is on a quiet section, and the forward is dropped at all on another
A neural network is an optimizer too, whatever. All this machine learning, including the in-house optimizer in the terminal
In given article in the beginning of topic suggested a simple optimizer through logit regression - very fast. You get a semblance of a walk-forward, i.e. a sliding optimization within a tester run. You can optimise this optimiser, anything
but you need to understand what you are doing and why))I'm just talking about the method itself in its bare form, which won't improve the TS much if there's no pattern per se
If you have reached the real world, the thing is only needed for automation, conceptually, not at all.
If you have reached the real, you only need it for automation, conceptually it's not needed at all.
If the pattern does not catch on, the VF is likely to show an approximate drain on the spread.
If you get to the real world, you only need it for automation, conceptually you don't need it at all.
Well yes, but maybe just fit more gracefully under all the chunks, and on new data still dick with butter
Once you start doing it, you realise that you can't reach the functionality of a tester - there are no ticks (not in the sense of no ticks at all, but in the sense that you'll get tired of modelling them). Therefore the solution is incomplete. Besides, each modification of the strategy will require a serious modification of the embedded optimizer. So we get a solution that is not all-purpose. That is why an idea arises - why not use the second terminal for automated optimization? And when you think carefully how to do it, you forget about it. What is the problem with manual launch of optimization once a week?
The problem is that the optimization must be performed for each pair separately (say, 60 pairs) and the best ones must be selected. The results also change when the broker or the account type changes and optimization must be performed again.
And all this takes a lot of time, taking into account that the optimization is carried out using real ticks, and the optimization in the clod has been cancelled using real ticks.
And if this robot is for sale and you don't know what broker the user is trading on, then self-optimisation will come in handy.