Optimisation and Out-of-Sample Testing. - page 9
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
Or better still, rewrite that bit of history altogether to make the councillor more comfortable. :))
Yeah and the future is better to write it yourself to know when to buy when to sell :)
>> yeah and you'd better write yourself so you know when to buy when to sell :)
"I have idiosyncrasies about rewriting after Mikhail Sergeyevich, GKChP (I happened to be on DB and BP at that idiotic moment), and Boris Nikolayevich - history is unpredictable :))))......... for forex)
As Mathemat says..... you try here, you try..... :( but all the same for yourself :)
No, really, if the issue with the test still somehow solvable, then with the optimization on these sites is a complete failure...... and, agree\ Granit77, that the screen, which I attached for you, on the trash does not pull )
And then you probably need to either remove a piece of history so it does not open this row of positions or add a piece of history so it can close them properly.
Yeah, I wish I knew what to remove beforehand :)
If you use Vita variant then you may get data fitting and I doubt that something good will come out of it.
No worse than "consistent" ..... I don't want to repeat myself - read his post about sets and subsets carefully.... - he's right ))
No guarantees either way......
As for the time consumption, I think it's more about the Expert Advisor's code or a lot of optimized data, rather than the optimizing loop code, which makes the optimization take a lot of time.
That's what I always deal with first of all, it's code optimization. It's fresh in my mind when my machine was feeble and I squeezed more out of it than on smart ones..... so the question is not about that.
Here's the thing: optimization is in progress and I'm requesting (let me see) 26195400 runs, in reality in the output (I've been optimizing for 3 years) about a thousand variants turns out. There is no way we can check this number manually. There is a "slice" left for forward "2008" in full. And here I can put these 1000 variants into "sequential" - let it discard 500, I don't mind :)
But when my machine asked for 26 million, it managed to do 5000 passes, and this 1000 without genetics, 24 hours it's asking :(((
>>agree Granit77, that screenshot I attached for you is not a dump.)
I don't pretend to be an adviser, it may not be a dumping ground. I'm just sharing my feelings: I've got a lot of drawdowns like that...
idiosyncrasy. Of course, nothing is thrown away, but the optimization result is marked as negative.
I don't pretend to be an adviser, it may not be a dumping ground. I'm just sharing my feelings: I'm not susceptible to such drawdowns.
Idiosyncrasy. Of course, nothing is thrown away, but the result of optimization is marked as negative.
:) change yourself (not yourself) )
seriously - is it "idiosyncrasy" or "idiosyncrasy" correct ?
:) change yourself (not yourself) )
I shouldn't have put a smiley face on it. I will change by all means, and it is in the direction of studying the subtleties of optimisation, so as not to be unsubstantiated.
So far I trusted my intuition, but the further I go, the more doubts I have.
P.S.
Mathemat has accustomed to speaking "you" to those with whom you find common ground. If you don't mind, let's keep it on a first name basis.
P.P.S.
Checked with the dictionary, the correct word is "idiosyncrasy",
Idiosyncrasy (from Greek ѭѭѭѭѭѭѭ - peculiar, special, unusual and ѭѭѭѭѭѭѭѭѭ - mixing), intolerance - a painful reaction that occurs in some ...
Well, happy Miner's Day!
Mathemat has made it a habit to speak "you" to those with whom you find common ground. If you don't mind, let's keep it on a first name basis.
I don't mind, although I don't speak chinese.)
.... Yeltsin remembered
Well, happy Miner's Day!
i can list a dozen other holidays..... do you need one? )))))
The question is not an easy one.
In general, until now I thought that first of all it is necessary to make a system that does not leak on a standard, invariable lot...... more ..... "that's how the standard one works", as few as.... but if you screw in a neat (I emphasize "neat" moniker, which does not distort its meaning, the result is completely different :)
They talk to us about it but we do not believe it :))
so 27 August was Miner's Day.
without the miners, Russia's first president would have had a different last name.
2.
the best mathematically sound money management is when starting from 10k at 0.1 lot and no more.
The topic should not have been hushed up. The question is still up in the air. After I've tried a lot of Expert Advisors which were getting much more profitable with "stupid" (one-pass) optimization, and then were losing if not on forwards, then on demo, if not on demo, then on real, I realized that until I solve this question for myself - I will not go further.... I've got nowhere :)
Almost a month is gone. Here is some preliminary result:
Strategy Tester Report
Tango-Mini
Alpari-Demo (Build 218)
Символ EURGBP (Euro vs Great Britain Pound )
Период 4 Часа (H4) 2005.07.04 00:00 - 2008.10.02 20:00 (2005.07.03 - 2008.10.03)
Модель Все тики (наиболее точный метод на основе всех наименьших доступных таймфреймов)
Параметры Counter=0; Filename="opt1.txt"; nameEA="Tango-Mini_EG240"; PauseBeforeTrade=11; MaxWaiting_sec=60; Dolya=0.5; Down=50; Lot=0.1; LotsMeneg=0; MinProfit=115; SumProfit=85; Tral=2; VF=2; StopLoss=225;
Баров в истории 6054 Смоделировано тиков 2905866 Качество моделирования 90.00%
Ошибки рассогласования графиков 0
Начальный депозит 1000000.00
Чистая прибыль 14620.93
Общая прибыль 32433.61 Общий убыток -17812.68
Прибыльность 1.82
Матожидание выигрыша 24.91
Абсолютная просадка 1096.96 Максимальная просадка 1824.99 (0.18%) Относительная просадка 0.18% (1824.99)
Всего сделок 587
Короткие позиции (% выигравших) 270 (79.26%)
Длинные позиции (% выигравших) 317 (75.71%)
Прибыльные сделки (% от всех) 454 (77.34%) Убыточные сделки (% от всех) 133 (22.66%)
Самая большая прибыльная сделка 415.05 убыточная сделка -442.03
Средняя прибыльная сделка 71.44 убыточная сделка -133.93
Максимальное количество непрерывных выигрышей (прибыль) 42 (2200.88) непрерывных проигрышей (убыток) 7 (-2822.10)
Максимальная непрерывная прибыль (число выигрышей) 2200.88 (42) непрерывный убыток (число проигрышей) -2822.10 (7)
Средний непрерывный выигрыш 6 непрерывный проигрыш 2
Дополнительно:
- максимум текущего убытка:- 1157 пунктов
- максимальное количество одновременно открытых позиций: - 21
I changed rules of entering and leaving on my own:
Strategy Tester Report
NACM
Alpari-Demo (Build 218)
USDJPY (US Dollar vs Japanese Yen)
Period 1 Hour (H1) 2005.07.04 00:00 - 2008.10.02 23:00 (2005.07.03 - 2008.10.03)
Model All ticks (the most accurate method based on all smallest available timeframes)
Parameters Counter=0; Filename="opt1.txt"; nameEA="NACM_UJ60"; PauseBeforeTrade=11; MaxWaiting_sec=60; Dolya=0.5; Down=50; Lot=0.1; LotsMeneg=0; x1=100; x2=148; x3=20; x4=111; MinProfit=44; Tral=0; VF=2; TakeProfit=73; StopLoss=296; Dopusk=7;
Bars in the history 21058; 6173610 ticks were simulated Quality 90.00%
Chart mismatch errors 0
Initial deposit 1000000.00
Net profit 4642.05 Total profit 12595.05 Total loss -7953.00
Profitability 1.58 Expectation of winning 16.52
Absolute drawdown 118.55 Maximum drawdown 1434.04 (0.14%) Relative drawdown 0.14% (1434.04)
Total trades 281 Short positions (% win) 155 (89.03%) Long positions (% win) 126 (90.48%)
Profitable trades (% of all) 252 (89.68%) Losses (% of all) 29 (10.32%)
Largest profitable trade 75.65 losing trade -306.86
Average profitable trade 49.98 losing trade -274.24
Maximum continuous wins (profit) 41 (1932.69) continuous losses (loss) 2 (-605.28)
Maximum continuous gains (number of wins) 1932.69 (41) continuous losses (number of losses) -605.28 (2)
Average continuous wins 10 continuous losses 1
Initial deposit 1000000 (?). During optimization I put a limit on the drawdown, there is a percentage there. If 30% from 10000 and from 15000 - these are values of different order, then 0,3% from a million, or a little more, approximately the same amount.
Yes, the ticks at the end of the charts are "close at stop"
Optimisation 24/6/6/3 - number of months.
24 is the main part, identifying sets of parameters which give acceptable results.
2 to 6 is OOS: run the resulting sets over these intervals.
Next, Excel and analysis:
- inadmissible small profit - scrap
- deals less than 70-80 on the main period - scrap
- negative balance on OOS - scrap
- gross mismatch between Profits and number of deals and periods - scrap
after that remain 10-15 lines....... for that the window in 3 months: just to see, if it is worth doing further choice - may be there are only minuses :)
If I see that it is worthwhile, then I group the sets - 2, a maximum of three and a final optimization across the interval ...... choose the best, and then the demo .... no guarantees, no statistics yet :)
..... and all this happens quite far from sets, which give the maximum profit on the main part of the optimization.
Thus, 90-95% of seemingly profitable combinations are rejected: Expert Advisor-Instrument-Timeframe.
Someone may say that it's not enough profit. But I "bet" on the reliability.
Criticism is welcome.
PS1 I do not pretend to authorship, most of what I used, described in this thread and on this forum.
PS2 :) sometimes it is not harmful at all to make a small mistake in a code which then will allow not on all ticks, and on control points to optimize......