Machine learning in trading: theory, models, practice and algo-trading - page 2069

New comment
 
mytarmailS:

I've got no one who earns money.)

So if you bought it, you earned it :)

 
mytarmailS:

Who has any ideas how to teach this automatically, I would be glad to hear about it?

no way

The problem with identifying patterns is very simple - they can be perfectly found only on the history,

but online they are detected 50/50 - i.e. it's a pattern or it's not our pattern that we expected

i have done a long ago a calculator, which first searches for signs (repeats/patterns) on the history and then it runs through it again without seeing any new bars - it doesn't even get 50/50, but approximately 10-15% of coincidences, and the rest of the combinations of bars never repeat, or repeat 1-2 times during the history

i really studied the euro, it may be the most non-technical currency ))))



SZY: search for "pattern" articles, there is a lot of material and new articles are written regularly... but only how to find a pattern on history ;)

 
mytarmailS:

You don't understand at all what I mean, there is no problem with identifying the pattern, there is a problem with how to describe the following reaction automatically

Here it is, the pattern, see?

the one in the blue rectangle...

And where was the reaction ??, look at the picture above!, no one did that, and no one wrote in any article...

You have to write it yourself - I have a type for such cases - Trident is called :)))

But, the problem is something else, they are rare, and because of this there is no training on them!
 

Ugh, how angry - identified indicators from the standard delivery, which give a different indicator at the same time, if they were run in different periods in the tester - I do not know how they are counted, but for machine learning they are dangerous!

iAO(Symbol(),TF);//Awesome Oscillator
iOBV(Symbol(),TF,Type_Vol);//On Balance Volume
 
mytarmailS:

You don't understand at all what I mean, there is no problem with identifying the pattern, there is a problem with how to describe the following reaction automatically

Here it is, the pattern, see?

the one in the blue rectangle...

And where was the reaction??, look at the picture above!, no one did that, and no one wrote that in any article...

I've been doing that since before the MO. The future turned out to be random, like yours (I've posted similar pictures here, too).
That's why I switched to MO, I thought it would be better at finding patterns.

I have not got any result yet.

But I still have hope.)

 
Aleksey Vyazmikin:

So I said, but the completeness will be small, ie out of a potential 1000 will be traded (found the class of "1") 100 of which 50 unprofitable.

30% - is a good result, but it is not always - now writing an article, there is a strategy by MA, it turns out on average 5% detect units, well, it's almost on the standard oscillators with default settings :)

Since 2019 the sample is out of training. It doesn't leak, and that's good.)

In my opinion this is not terrible. It's just unidentified entry options. Missed profit. And maybe not profit, because in addition to these 900 missed inputs, could add 2000 unprofitable. And the system cannot separate them from each other, so it skips them.
The main thing is that these remaining 100 out of 1000 are profitable.

 
elibrarius:

In my opinion this is not terrible. It's just unidentified entry options. Missed profit. And maybe not profit, because in addition to these 900 missed inputs, you could add 2000 unprofitable ones. And the system cannot separate them from each other, so it skips them.
The most important thing is that these remaining 100 out of 1000 were profitable.

I have a set where 11 trades out of 1000 is very little, at the same time I can make a quick 100 with my hands in excel!

So there must be methods - you have to think.
 
Ehhh... I'm in that boat myself... unfortunately...
 
Aleksey Vyazmikin:

I have a set where 11 deals out of 1000 is very little, at the same time I can quickly make 100 with my hands in excel!

So methods must be - we have to think.
Something she lacks to separate flies (0) from cutlets (1)))
 
elibrarius:
Something she lacks to separate flies (0) from cutlets (1) )))

There's only 24% of units, and the strategy is profitable for 7 years on minutes!

I cannot use CatBoost, it has all zeros, only genetic algorithm on R somehow catches the logic, but poorly.

1...206220632064206520662067206820692070207120722073207420752076...3573
New comment