Question on genetic optimisation - page 8
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
This is where you dig. For example, for the presence of graphic objects in use. Bar control. What else is there...
P.S. If there are indicators -- check each one too. It is possible that some of them are drawing.
I don't use graphical objects, all calculations are performed on the zero bar only and reset to the shift register, they are corrected only by opening prices, I don't work with history, thus re-rendering is excluded. All indicators are built in the system and intended for working only with the current values of signals because there are no cycles in int start(). The whole system in this regard has been optimized to the maximum, nothing unnecessary. In the visualization mode of the indicator, I see the correspondence of the strategy, after the test is stopped the lines of the indicator, which was used by the EA, and the indicator, which is additionally attached to the chart for visualization in stepwise mode, are identical.
The question is not about strategies, I have revised about a dozen of them during this month, they are robust, the question is their improvement through parameter optimization, or rather the selection of the optimal variant, there is a deadlock, the optimization does not add stability and vice versa and no assurance that the "good" parameters will work in real trading.
I don't use any graphical objects, all calculations are made only on the zero bar and reset to the shift register, I don't work with history, thus no redrawing is possible.
>> Well, well.
Well, well.
And how do you decipher that?
I'm amazed at your self-confidence. Answer the question: why do you need a zero bar? That's the first problem: the operation of your TS with an already formed bar in the tester and the operation with the forming bar in real time will be very different.
Oh, mate, you'll have to make up your own mind, won't you? You're writing, aren't you?
Angela писал(а) >>
I don't use graphical objects, all calculations go only on zero bar and are reset to the shift register, I don't work with history, thus, redrawing is excluded.
Oh, mate, you'll have to make up your own mind, won't you? I mean, you write:
I guess we don't speak the same language.
I think the problem was sucked out of my hand :-))
it's pretty simple - Total deals - 44 ...
that's not even enough to predict the weather :-))
hence the discrepancy in the results
>> Again, the question is not about strategies, I've revised a dozen of them during this month, they are robust, the question of their improvement through parameter optimization, or rather the best option, here is a deadlock, the optimization does not add stability, and vice versa, and no assurance that the resulting optimized "good" parameters will work in real life.
Everything is correct. But unfortunately the optimizer has no idea about your TS.
At some point, using dumb "genetic" brute force method, he jumps from your TS to something else.
Well imagine that an aperiodic process (eg exponent) and oscillatory (sine wave) are described by the same equation.
Only the coefficients are different. You optimise all the coefficients at the same time, the aim is to reach some point faster.
And suddenly, instead of a monotonically increasing value you get an oscillating process. But you reach this point the fastest.
Only this is not what you need. And you cannot tell a "clueless" computer - well, I don't want a sine wave, I want an exponent - there is no such button in the optimizer.
Please advise, who knows, how much do the chart mismatch errors distort the test result, can we trust such a test?