Machine learning in trading: theory, models, practice and algo-trading - page 438
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
It's all one formula, I just wrote it in three steps to make it clearer. So the signs will not be a problem, because there is a squaring.
((a5-b5)+(b4-a4))^2 = (delta 5 + ( - delta 4))^2
I.e. you will still underestimate the error for delta sequence 0, +10, +15,+12, +5
That's a cool processor you have! ))
Mine adds and subtracts faster than multiplication, and modulus is found just by equating bit 64 to zero.
Don't be lazy to do the test, my multiplication and difference differ by 5%(*>-) but the module weighs as much as half of the multiplication, in the end your version is slower than mine by 50%
If difference without modules it's a bit faster (~3%), but module slows down 1.5 timesUnfortunately, the level of the community over the past 10 years has not gone far from the AI of the late Yury Reshetov, as they say "what was, is, and there is nothing new under the sun.
Checked. Expert Code:
int OnInit()
{
double res=0;
double p[];
int bl=CopyClose (_Symbol,PERIOD_M1,0,2000,p );
uint t1=GetTickCount();
for(int i=0;i<2000;i++){
for(int j=0;j<2000;j++){
for(int k=0;k<2000;k++){
res+=p[i]*p[j]+p[j]*p[k]+p[k]*p[i];
if (res>1000000000){res=0;}
}}}
uint t2=GetTickCount(); Print("1 Calc time=",t2-t1);
t1=GetTickCount();
for(int i=0;i<2000;i++){
for(int j=0;j<2000;j++){
for(int k=0;k<2000;k++){
res+=MathAbs(p[i]-p[j])+MathAbs(p[j]-p[k])+MathAbs(p[k]-p[i]);
if (res>1000000000){res=0;}
}}}
t2=GetTickCount(); Print("2 Calc time=",t2-t1);
return(1);
}
void OnTick() { return; }
Results of several runs
2017.07.01 20:50:57.268 test (EURUSD,M1) 1 Calc time=5897
2017.07.01 20:51:02.227 test (EURUSD,M1) 2 Calc time=4961
2017.07.01 20:51:14.359 test (EURUSD,M1) 1 Calc time=5913
2017.07.01 20:51:19.290 test (EURUSD,M1) 2 Calc time=4929
2017.07.01 20:51:58.296 test (EURUSD,M1) 1 Calc time=5960
2017.07.01 20:52:03.357 test (EURUSD,M1) 2 Calc time=5070
2017.07.01 20:52:21.364 test (EURUSD,M1) 1 Calc time=5928
2017.07.01 20:52:26.303 test (EURUSD,M1) 2 Calc time=4930
Your variant is 20% slower. Honestly, I didn't expect multiplication to count so fast.
PS.
Removed MathAbs to see if it's really that slow:
2017.07.01 21:06:03.844 test (EURUSD,M1) 1 Calc time=5943
2017.07.01 21:06:08.793 test (EURUSD,M1) 2 Calc time=4945
The speed has not added, so MathAbs are still counting very fast, it is logical to do it by zeroing the 64th bit, and bit operations, very fast....
Unfortunately, the level of the community over the past 10 years has not gone far from the AI of the late Yury Reshetov, as they say "what was, will be, and there is nothing new under the sun
So it must be... very sad
It is normal. The problem is that people on the forum are constantly changing and, having found their way out of the forum, they either leave trading because of the hopelessness of the event.
At present I seem to have found the solution to the problem of constructing the TS on multilayer (more than two)) And although I haven't even started the TS yet, I limit myself to
Although I haven't even started with TC itself, I limit myself to experiments with time series and statistics, but, generally speaking, I have nothing special to read and write about on the forum. I haven't visited the forum for almost a week.
This is normal. People on the forum are constantly changing, and finding their own outgrow the forum, or quit trading for the hopelessness of the event.
I, for instance, now seem to have found the solution to the problem of building TS on multilayer (more than two)) And although I haven't even started the TS yet, I limit myself to
Although I haven't even started with TC itself, I limit myself to experiments with time series and statistics, but, generally speaking, I have nothing special to read and write about on the forum. I haven't visited it for almost a week.
I'm bored already for a year and a half, only this topic is interesting and some other articles, the rest seems to be kindergarten already.)
It takes a long time to explain, in general ZZ is not suitable. I am telling you that for sure.
I do not understand what I mean. My Expert Advisor improves its performance when it uses a phase (analyzes structure and its vector), but it is not suitable for training - it is strange.
We must decide what we want to do. Predict or categorize. The approaches are absolutely different, but they have the same purpose....
I'm interested to classify now, i.e. to apply filters under certain conditions that I can't identify myself - sometimes filtering happens and I suppose that NS can solve this problem.
About EZ - I do not understand, my Expert Advisor improves its performance when using EZ (analyzes the structure and its vector), but it is not suitable for training - it is strange.
They just do not know how to prepare it.)
I am now interested in classification, i.e. to apply filters under certain conditions, which I myself can not identify - sometimes filtering happens, and I believe that NS can solve this problem.
NS can, but here is the same issue of preparation, and probably an issue of TC application ideology. Let's say specific applications pose the question quite differently.
It seems to me that this is how they should be prepared). I.e., NS should be an addition to TC and solve its own problems.
Unfortunately the level of the community over the past 10 years has not gone far from the AI late Yuri Reshetov, as they say "what was what will be and there is nothing new under the sun.
I do not understand, Jura no longer?