Machine learning in trading: theory, models, practice and algo-trading - page 2958

 
Vladimir Perervenko #:

Of course I realised that. But do the people using the example get it?

52 per cent efficiency. They should understand.

MAE=0.0005

 
Let me explain the situation a little bit. It is clear to everyone here that the examples are purely technical in nature. But people who are just getting acquainted with MT and have come from data science may have an unpleasant impression when they see this example. That is why Vladimir Perervenko suggests to calculate all normalisation using training data, so that the information does not leak out of the test sample.
 
Aleksey Nikolayev #:

This problem is solved by using an appropriate loss function when training MO models. There are two problems associated with this. The first one is technical - standard loss functions in MO packages are related to profit maximisation only indirectly, which leads to the necessity to create custom ones. This is quite difficult to do - you need to have a good understanding of MO packages at the code level. If this problem is solved, then the second, mathematical problem may appear, when the custom loss function is not good for training the model.

If someone solves such a difficult problem, they are unlikely to share the solution.

Apparently, functions are worth reading about loss functions. If I understand correctly, it turns out, for example, in a market where you can only buy the loss function:

sum of the difference between the buy and sell prices, for cases where: "buying price" - "selling price" < 0.

In this case we get a sum with a negative value, i.e. it should tend to 0 or to the maximum. To make it tend to the minimum before the sum of differences we need to put a minus sign.

Regarding the fact that if someone has solved such a difficult problem will not share the solution, I did not count on it, I just want to understand how to maximise profit or to see as an example of at least some model (it does not matter whether it is profitable or not, just to understand the general framework).

And so, sure, I think custom loss functions will generally be bad for training a model because that's not the main part of it.

 
Maxim Dmitrievsky #:
Optimise your strategy (if you are too lazy to make another markup) by the maximal profit criterion, then train on this TS. Or take any profitable TS from the market. The same training with a teacher.

If you are interested in deriving a TS on the basis of NS only, I can offer a variant from my last article. You can do it in a similar way. I was initially wondering how to do such a thing. Exclusive.

I would not say that I am lazy to do any other markups, I try different variants and since I am not a senor in machine learning, when some idea comes into my head, I try to find at least some variants of examples with attempts to achieve the result.

When I tried to make a parametric variant of the solution with its own values of indicators, but it turned out that there are so many variants of the set of values of indicators that with the current computing power selection of parameters will be carried out almost 10 years)

I was surprised when I read the phrase "take any profitable TS from the market". I didn't even consider such an option, as I thought they were not there.

 
mytarmailS #:
I've shared so many times already that at some point I got sick of it.....

People start thinking and talking about what I discussed years ago, but no one understands.

Not everyone was familiar with machine learning years ago. If your posts are re-read, the discussions should be found in the same thread/branch?

Apparently it's worth trying to make sense of all 3000 pages of the thread while it's still growing.

 
Elvin Nasirov #:
Apparently, you should try to understand all 3000 pages of the topic before it grows.

It is better to understand the articles by Vladimir Perervenko and Maxim Dmitrievsky. You will spend your time more usefully. The forum is mostly water.

Vladimir Perervenko
Vladimir Perervenko
  • www.mql5.com
Профиль трейдера
 
Forester #:

Better understand the articles by Vladimir Perervenko and Maxim Dmitrievsky. You will spend your time more usefully. The forum is mostly water.

Thank you!

 
Elvin Nasirov #:

Not everyone was familiar with machine learning years ago. If your posts are reread, the discussions should be found in the same thread/branch?

Apparently it's worth trying to make sense of all 3000 pages of the thread while it's still growing.

Your problem is an optimisation problem , searching for unknown parameters

Here is the ONLY article you need to study https://www.mql5.com/ru/articles/2225.


If you want to teach AMO to maximise profit and minimise drawdown:


you need to

1) create a fitness function, a function that will count profits and losses from trading signals.

2) any AMO algorithm that will generate signals for trading, for the fitness function (p.1)

3) any optimisation algorithm (genetic, particle swarm, churn) - which will generate signals as targets for AMO (item 2).


The algorithm is as follows

1) AO creates a target for AMO

2) AMO learns on this targeting

3) AMO creates a forecast of trade signals

4) trade signals are evaluated by FF and produce a result

5) the result of FF is evaluated by AO and maximised/minimised further and so on in a circle until you get an acceptable result.


==========

AO - optimisation algorithm

AMO - machine learning algorithm

FF - fitness function

=========


ps. if you want to work with neuronka and not with any AMO, you can change weights using AO tools, without learning the targeting.

Самооптимизация экспертов: Эволюционные и генетические алгоритмы
Самооптимизация экспертов: Эволюционные и генетические алгоритмы
  • www.mql5.com
В статье будут рассмотрены основные принципы, заложенные в эволюционных алгоритмах, их разновидности и особенности. На примере простого эксперта с помощью экспериментов покажем, что может дать нашей торговой системе использование оптимизации. Рассмотрим программные пакеты, реализующие генетические, эволюционные и другие виды оптимизации и приведем примеры применения при оптимизации набора предикторов и оптимизации параметров торговой системы.
 
mytarmailS #:

Your problem is an optimisation problem, searching for unknown parameters.

Here is the ONLY article you need to study https://www.mql5.com/ru/articles/2225


if you want to teach the network to maximise profit and minimise drawdown:


you need

1) create a fitness function, a function that will count profits and losses from trading signals.

2) any MO algorithm that will generate signals for trading, for fitness function (p.1)

3) any optimisation algorithm (genetic, particle swarm, churn) - which will generate signals as targets for AMO (p.2).


algorithm like this

1) AO creates a target for AMO



AO - optimisation algorithm

AMO - machine learning algorithm

FF - fitness function

Can you give me any example of an AO? I was under the impression that finding a teacher is piece work and lends itself to automation.

 
СанСаныч Фоменко #:

Can you give me any example of an AO? I was under the impression that finding a teacher was a piece job and lends itself to automation.

You can--

You need to make it as simple and reproducible as possible, if there is interest...

It would take a little bit of time.