Mashki and I. Captured by illusion... - 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
---- это не шутка. отчёт прикладываю.
вот теперь стоит потестить стратегию на тесторе. ;)
Interesting.
I think the results will be worse - without looking to the 'future'.
2) If there is an algorithm (however crude) for marking points before turning, (maybe even looking into the future), we would now like to see differences in the behaviour of deviations. - Please do so, it will allow us to move on.
4) if we manage to split the set, the four slices will be even more informative.
---
5) The statistical "representativeness" of the material only makes sense to discuss if there are suggestions about the necessary level of these representations. I have given all the parameters in the illustrations, including N - sample size.
--------------
6) But your assertion about the impossibility of splitting the set puts me in an awkward position.
And that - that even trying to do so contradicts the very meaning of the idea.
I don't know what to think.
How to solve "homework"?
6) You must have misunderstood me. There are only 4 pictures, each of them refers to a different object. The point of the idea is to separate states with different contexts. In your particular case - using the distribution of price deviations from the MA.
You give a picture that is generated for the whole story. In it all the contexts are mixed together and averaged. And so from 4 different sides. What next ?
Here (and the next 2 posts) it was said on this subject https://www.mql5.com/ru/forum/123154, but there was no response from you.
Try building a distribution for the window. If you want to play on H1, then, with an average hold time of 5 hours, on minutes it would be a 300 window and on M5 a 120 window. 300 is better, but 120 is also a bit of a stretch.
Further, because the distribution deviation from the normal distribution is not expressed by a single number, it cannot be treated as a normal parameter. All you need to do is to manually draw such distributions for pivot regions and trend regions and try to classify them by your own eyes. And then try to come up with formal criteria of their recognition and separation. And then write code and collect statistics of these recognitions.
IMHO, you are too complicated "homework".
2) If you don't have your own I/O algorithm, then use zigzag vertices. Choose the parameter ZZ at the rate of 2*TR.
5) The sample size in this case is the size of the window. This is what gives the data set to build the distribution. And, of course, it makes sense to choose it so that the distribution has a more or less clear form, but also not too big to reduce the lag. It only makes sense to total the entire sequence to solve the partitioning problem when this partitioning is already done, thinly or thinly. Then the recognition success/failure statistics will also appear.
4) The set can only be divided (if it is possible) when comparing images with the same tool, and not with different ones. And for that you need quite a lot of these pictures, and they must represent all the contexts. Then it will become clear whether you can classify them on the basis of these pictures.
4) Множество удастся разделить (если удастся) только при сравнении картинок с одним инструментом, а не с разными. И для этого нужно достаточно много этих картинок, и они должны представлять все контексты. Тогда и станет ясно сможете ли вы их классифицировать на основании этих картинок.
Thank you...
But you should look at the pictures in the attachment...
You can see two states there, according to the "crossing of the masks" splitting algorithm. As Alexey said ;)
I'm afraid to dig further.
6) Наверное вы меня не поняли.
Вы приводите картинку, которая сформирована для всей истории. В ней все контексты смешаны в кучу и усреднены. И так с 4-х разных сторон. Что дальше ?
Have you seen the report?
There's a lot to compare it to after splitting.
it takes a "pile" to define "not a pile". (c) Gardener
for those who don't like "pictures" and "excuses".
txt replace with htm !
:)
But you should look at the pictures in the attachment...
You can see two states there, according to the "overlapping of the masks" separation algorithm. As Alexey said ;)
I'm afraid to dig further.
Is it in the attachment on the previous page ?
I really didn't see it.
What are CZZ and FC ?
What is on the upper left diagram and the two it is divided into ? Principle of division ?
In what units are the axes ?
So, what's the bottom line?
Proof that in a series of 100,000 observations the "intersecting" wagons and zigzags produce 14% of signals that are profitable in perfect output?
Where are the inquisitive researchers, fierce Gardener opponents and just traders?
;)
Isn't a simple algorithm
1) Ищем по модулю расстояния между машками MaV_MaS. (200 и 15)
2) если оно меньше 10 пунктов, и последняя вершина 33 выше текущей цены не менее 25 пунктов - смело продавайте.
если последняя вершина 33 ниже текущей цены не менее 25 пунктов - смело покупайте.
Стоплоссы = 125 пунктов, ТР =600 пунктов ;) Оптимальнее в случае противоположного сигнала закрывать.
inspired anyone?
Zigzag with parameters (from the script) 20,1,3.
M5 timeframe eurodollar.
;)
Что такое CZZ и FC ?
Что на верхней левой диаграмме и на тех двух на которые она разделяется ? Принцип разделения ?
В каких единицах оси ?
If you're too lazy to read all the posts, see code -
FileWrite(hFile,"date","Time",
"<O>","<C>","<L>","<H>","<V>",
"DirectH","Forecast","ZZ","Direct",
" F-ZZ"," |F-<C>|"," |<C>-ZZ|","maV100","maS33", "МаV_O",
"MaS_O","MaV_MaS","dVaR","Std","Dstd"
);
for (i=Records;i>=0;i--)
{
ZZ=iCustom(NameVal,Tf1,"ZigZag",20,1,3,0,i);
if (ZZ!=0)
{
bS=!bS;
if (!HTF && MathAbs(ZZ-aForecast[i])>Point)
ZZ0=ZZ;
else if (MathAbs(ZZ-aForecast[i])<Point)
ZZ0=aForecast[i];
}
else if (i==Records) {Records--;continue;}
if (aForecast[i]>=10000) {break;}
aZZ=ZZ0;
aVol=iMA(NameVal,Tf1,15,0,0,1,i);
aSpeed=iMA(NameVal,Tf1,200,0,0,1,i);
FileWrite(hFile,dataConv(iTime(NameVal,Tf1,i)),
iOpen(NameVal,Tf1,i),iClose(NameVal,Tf1,i),
iLow(NameVal,Tf1,i),iHigh(NameVal,Tf1,i),
iVolume(NameVal,Tf1,i),aS[i],
aForecast[i],aZZ,bS
, (aForecast[i]-aZZ),
(aForecast[i]-iClose(NameVal,Tf1,i)),
(aZZ-iClose(NameVal,Tf1,i)),
aVol,aSpeed,
aVol-iOpen(NameVal,Tf1,i),
aSpeed-iOpen(NameVal,Tf1,i),
aVol-aSpeed,
iCustom(NameVal,Tf1,"D_Var",N1+120,15,7,2,false,0,i),
iStdDev(NameVal,Tf1,15,0,0,1,i),
iStdDev(NameVal,Tf1,15,0,0,1,i+1)-iStdDev(NameVal,Tf1,15,0,0,1,i)
COMPUTE filter_$=(Abs(MaV_MaS)<=0.001 & CZZ>0.0025). Context B
COMPUTE filter_$=(Abs(MaV_MaS)<=0.001 & CZZ<-0.0025). Context S
points and frequency. Normal frequency graph. :)
I'm personally shocked. Fucking statistics, science!
Or is it self-deception?
Я лично в шоке. Статистика блин, наука!
Или самообман?
Statistics is precisely self-defeating. Science in this case is contextualisation (i.e. clarification of conditions).
The rabid opponents are stewing in the corners.