a trading strategy based on Elliott Wave Theory - page 68

New comment
 
<br / translate="no">Rosh 05.07.06 14:57

14:54:05 ChannelStDev3 EURJPY,M15: Found 824 channels satisfying the criterion for 1000 bars


And what criterion do so many channels meet?
 
Rosh
... Object handling is time-consuming (almost a third of the non-optimised version) - drawing during the backtest is undesirable. Although
I've always suspected that objects are a heavy thing, so I tried to use indicator arrays for visualization. By any chance, did you compare the speed of one and the same algorithm when implemented as a script and as an indicator, for example? In general, visualization on the history is extremely desirable for any algorithm.
I do not quite understand optimism as well. There are point characteristics - they can be written into arrays, and there are channel characteristics - they must be fully calculated every time. In principle, recurrence schemes are possible, but if it is more or less obvious for points, then for channels. I will have to think about it.
 
In principle the calculations have been justified, on a sample of 1000 bars the algorithm is about 500 times faster.
I ran it on 3000 bars, it is 300 times faster, still not bad.
2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: removed<br / translate="no"> 2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: deinitialized
2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: Execute deinit()
2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: a=0.0057 b=146.754 lastBar1 firstBar=46 StDev=0.0998
2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: Found 2831 channels meeting criterion for 3000 bars
2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: They are in 6 series
2006.07.05 15:11:40 ChannelStDev3 EURJPY,M15: Time of the normal algorithm is 5094 ms
2006.07.05 15:11:35 ChannelStDev3 EURJPY,M15: Time of optimized algorithm 16 ms
2006.07.05 15:11:35 ChannelStDev3 EURJPY,M15: lastBar=1
2006.07.05 15:11:35 PM ChannelStDev3 EURJPY,M15: initialized
2006.07.05 15:11:29 ChannelStStDev3 EURJPY,M15: loaded successfully
 

Rosh 05.07.06 14:57

14:54:05 ChannelStDev3 EURJPY,M15: Найдено 824 каналов, удовлетворяющих критерию, на протяжении 1000 баров


And what is the criterion for satisfying so many channels?


The simplest one is the RMS of two thirds > the RMS of the whole sample.
 
Rosh
... работа с объектами отъедает значительное время(почти треть неоптимизированного варианта) - рисовать при бек-тесте нежелательно. Хотя
I always suspected that objects are a heavy thing, so I tried to use indicator arrays for visualization. By any chance, have you compared the speed of one and the same algorithm when implemented as a script and as an indicator, for example? In general, visualization on the history is extremely desirable for any algorithm.
I do not quite understand optimism as well. There are point characteristics - they can be written into arrays, and there are channel characteristics - they must be fully calculated every time. In principle, recurrence schemes are possible, but if it is more or less obvious for points, than for channels. I will have to think about it.


Indicators vs. Expert Advisors. The developers have repeatedly said that each type is executed in its interface thread, although not everyone must remember the priorities of these threads.
But it is better to see once than hear 100 times :)
Take the codes from this article - http://www.alpari-idc.ru/ru/experts/articles/20.html - and run the EA on a single pair on NFP (especially since there will be soon).
Warning! if you are planning to trade at this point in time - you should not hang the EA!!!

The question of priorities between the indicator and the EA will fall away :)
 
Honestly, I don't know how you manage not to count anything extra? I don't seem to be counting anything unnecessary. And how do you manage to reduce the calculation by a hundred times? Probably, if this is done as a separate article, it will be interesting to all in terms of language algorithm. After all, we still need to add much to channel search to get a working Expert Advisor. Although, I think that it is possible to explain the algorithm here and without putting the code itself - everything becomes clear to me anyway.

ZS: What is NFP?
 
solandr
ZS: What is NFP?

Non-Farm Payrolls (the number of people in employment, excluding agricultural industries)
One of the most important indicators, it shows the change in employment in the country. There is an opinion that a 200K change in this index can be equated to a 3% increase in GDP. It is published, as a rule, on the first Friday of each month. The release often causes sharp movements in the market. The nearest NFP is this Friday, 16:30 MSK
 
Honestly, I don't know how you manage not to count anything extra? I don't seem to be counting anything unnecessary. And how do you manage to reduce the calculation by a hundred times? Probably, if this is done as a separate article, it will be interesting to all in terms of language algorithm. After all, we still need to add much to channel search to get a working Expert Advisor. Although I suppose the algorithm can be explained here without pasting the code itself. <br/ translate="no">


"Think, Stirilitz, think" :)

I also do not understand some things - I have not even been able to start writing the Expert Advisor, and you have been testing it for almost a month already :)
If you don't solve the problem - I will send you an email. The fact that the problem has a solution is worth a lot. Because knowing that an EA using Vyacheslav's method has a positive expected payoff is not comparable to knowing the coding algorithms :)
Knowledge is primary, coding skills are secondary.
 
Rosh:
Indicators vs Expert Advisors...

That is, we can assume that Expert Advisors (scripts) have a higher priority in real time. However, will it make the same difference for the tester?
solandr:
Of course there is no doubt that we would like to get a faster calculation algorithm especially for history tests, but on the other hand this methodology does not require multimillion passes on the tester

Nevertheless, you are working on the tester. And I agree with you on that. For example, as I understand it, the probabilities are taken for a normal error distribution. Which in this case is incorrect. An idea of the difference between the real error distribution and the normal one can only be obtained from history. But, I'm afraid, only an idea, it is the real error distribution that can turn out to be the most volatile parameter. By the way, here is an example from another instructive indicator:

In the middle of the image we can see quite a stable channel but we have a feeling that movement from the centre to the upper border has quite a different price than movement to the bottom. Which subsequently justifies itself :) .
 
<br / translate="no">.


Rosh 05.07.06 14:57

14:54:05 ChannelStDev3 EURJPY,M15: Found 824 channels satisfying the criterion for 1000 bars


And what criterion do so many channels meet?


The simplest one is the RMS of two thirds > the RMS of the whole sample.


Interesting at about 4000 bars only 180 met this condition + that for the last 1/3 did not fall out of the 99% interval
1...616263646566676869707172737475...309
New comment