Machine learning in trading: theory, models, practice and algo-trading - page 231

 
mytarmailS:

I personally did not even understand anything )

But I'm allowed, I don't consider myself an expert.

The system is complex. In the end it realizes the full automation of the process of adjusting the input parameters, effectively adapting to the changing market dynamics. That's the point of it.
 
mytarmailS:

Personally, I didn't even understand anything.)

You are not alone, for the collective that "understood nothing" includes the author himself.

And that's the point of everythingReTeg Konow writes.

The goal is to destroy the branch and the easiest way to do this is to use texts that have no content at all.

The goal is achieved. The thread not only doesn't fit the title - it has ceased to have any content at all.

 
SanSanych Fomenko:

You are not alone, for the collective that "understood nothing" includes the author himself.

And that's the point of everythingReTeg Konow writes.

The goal is to destroy the branch and the easiest way to do this is with texts that have no content at all.

The goal is achieved. The thread not only doesn't fit the title - it has ceased to have any content at all.

Well, that's trolling. You do not understand - so ask, I will explain in detail. Just ask a specific question.
 
Retag Konow:
The system is complex. Ultimately, it implements full automation of the input parameter tuning process, effectively adapting to changing market dynamics. That's the point of it.

Yes, that is the point.

Technically, any control structure with adaptivity (feedback with quality/loss functionality) can be attributed to MO, for example, if you take an ordinary machine, and every N bars to search for its optimal parameters for the previous N bars, by a simple grid, it will also be MO. The essence of MO is in accumulation of "experience", in transformation of data into a model.

PS: In the above "perseptron" more precisely... Poorly programmed simple scalar product, lacks main thing, ALGORITHM OF TRAINING, there this important business is delegated to MT optimizer and MO is exactly double model - optimization algorithm, not only model, and not every optimization algorithm will work for every model, for example MLP cannot be optimized by grid or genetics, backprops are needed and so on.

 
Transcendreamer:

in the same way, in forex, a bot can successfully pass one segment/level and fail another

because he has no trader's intelligence and the bot "doesn't understand" what he's doing...

And what is trader's intelligence?

In simple terms, this is trading experience, i.e. a trader has 1) - certain patterns of market behavior (market patterns) and 2) - certain patterns of behavior in this pattern (patterns of behavior in the market pattern).

This can be programmed, can't it?

Moreover, searching for these good patterns can be programmed, can't it?

transcendreamer:

..

In order to be successful at any area/level, the bot must have an object model of the world in which it exists

that is, the algorithm must not simply optimize the patterns

The algorithm must operate with semantic categories and describe the situation as it is seen by the trader/gamer

The bot must distinguish the types of objects and their characteristics and evaluate the danger of the situation dynamically

And this requires quite a different level of heuristics than the simple neural network optimization

The result of learning must be a semantic model and knowledge about objects and processes

If it's not there trade bots are doomed to random poking

I agree completely, the grail is not in the MO, the grail is in these little cubes, even in the case of mario they do it, not like with the non-stationary market

й

ф

It's called data preprocessing, even in this thread no one is doing it.

This is exactly what the MO sees, a concentrate of reality cleansed of noise.

If to do it and do it adequately, then any optical analyzer will be able to trade as good as a human, or even better.

 

Yes, that is the point.

Technically, any control structure with adaptivity (feedback with quality/loss functionality) can be attributed to MO, for example, if you take an ordinary machine, and every N bars to search for its optimal parameters for the previous N bars, by a simple grid, it will also be MO. The essence of MO is in accumulation of "experience", in transformation of data into a model.

In other words, do you agree that it is possible to achieve results of the standard MO approach with neural networks using the non-standard MO approach (since you have recognized this approach as such)?
 
toxic:

MoD is first and foremost an engineering art, the result justifies any concepts. Give me the result. Here's a challenge for you: https://numer.ai/

As far as I'm concerned, any member of the branch who considers himself useful to this branch is just obliged to show the result in this thing )
 
Konow's tag:
In other words, do you agree that it is also possible to achieve the results of the standard MO approach with neural networks using a non-standard MO approach (since you have recognized this approach as such)?
Surprise me and the other participants))
 
toxic:

Yes, that's exactly the point.

...

The essence of MO is the accumulation of "experience," the transformation of data into a model.

This conclusion confirms that my concept may have real value.

In any case, I will certainly get to work on it. When the time comes.

 
Combinator:
As for me any member of the branch, who considers himself useful for this branch, just have to show the result in this thing )

I agree, in my opinion, if a person can at least just run the data and get the logos below 0.69300 (random) then has the right to talk about AI and MO here, the rest are not profitable

My result https://numer.ai/ai/toxic