What to feed to the input of the neural network? Your ideas... - page 83

[Aleksei Kuznetsov]

Рaра Нoth #:
I did not do it, but I saw the result with my own eyes and the whole essence immediately grasped, believe or not your problems.

If you have seen a profitable thing and understand how it works, why don't you do it? Or are you not sure about something (the result or the essence) to spend your time on it?

[Arch]

Forester #:

If you have seen a profitable thing and understand how it works, why don't you do it yourself? Or are you not sure about something (the result or the essence) to spend your time on it?

I've already made mine, the presentation is on the wall. I don't see the point in doing anything else, much less going through the neuronics

[Arch]

Maxim Dmitrievsky #:
Can you give me the full text of the interrogation, please?

It's about the same with everyone, they fade out immediately if you tell the truth.

[[Deleted]]

Рaра Нoth #:
Pretty much the same with everyone, fades out immediately if you tell the truth

😁

[Aleksey Nikolayev]

Рaра Нoth #:
It's about the same with all of them. They fade out immediately if you tell the truth.

The subject of kurva-fitting has not been broached.)

[Ivan Butko]

I came up with an idea, I asked the chat, he told me: "bullshit, fitting is the worst, will not work, think further man".

And described the arguments against: that kven, that dipsic. Gpt dryly expressed nothing.


The point:

What if the target is the result of running the weights as an independent TS?

Maybe someone who has tried and give a brief characteristic of the method.


For example:

1) Initiate the weights.

2) Run the test

3) Choose a target from the trading result in such a way that it shows the shortcoming of the system in the form of a value that will play the role of error: drawdown, inverted recovery factor (when less than 1 - this is the maximum error, for example, by converting through the activation function, so that if the PV increased to 10, it would look like striving to 0 - reducing the error).

4) Train the NS according to the error.


You will say: "This is pure RL".
But, in RL there are a lot of actions and there is a different formula.

And here there is nothing at all from the RL mechanism, except the similar essence. No state table, no pending awards, no policies, no endless date-sets with targets, etc.

[Сергей Криушин]

A person sees what he only wants to see or what he is trained to extract and see, often the visible things are not perceived at all, because there is no knowledge that there is such a phenomenon or regularity, we drive ourselves into the framework of commonness and far-fetched regularity, which does not exist at all, what regularity can there be in the Brownian motion of a quote on a chart, as a particle, which is influenced by the multiplicity of all factors, there is one regularity for sure - fractal, but it cannot grow higher than the sky and sometimes merges, inflated bubbles burst - corrected, and, as it seems to me, only an intuitive feeling of temporary enlightenment can have any meaning - "eureka!!!!" sort of

[Ivan Butko]

Сергей Криушин #:
A person sees what he only wants to see or what he is trained to extract and see, often the visible things are not perceived at all, because there is no knowledge that there is such a phenomenon or regularity, we drive ourselves into the framework of commonness and far-fetched regularity, which does not exist at all, what regularity can there be in the Brownian motion of a quote on a chart, as a particle, which is influenced by the multiplicity of all factors, there is one regularity for sure - fractal, but it cannot grow higher than the sky and sometimes merges, inflated bubbles burst - corrected, and, as it seems to me, only an intuitive feeling of temporary enlightenment can have some meaning - "eureka!!!!" sort of thing.

It's handwringers. Three different people trade on one subjective-analytical system with patterns and scenarios of behaviour.

No grids, no martins, no mashes, no standard indicators.

Purely trained by eye on the chart, trained on the context (when you take into account the totality of factors). And there are many of them.


It all works, but not everyone can do it.

I have a hard time with semi-automatic systems, where I have to think with my head, I am too used to formalisations, digging into Expert Advisors.


But there are working systems and this is the motivation - to make NS/II/machine understand the nuances of working with probability theory, using both short stops in time, not in a hurry to close, and the right time to be out of the market and wait for ideal conditions.

[[Deleted]]

Ivan, input a question to the neural network: what is more profitable (profit/risk) and stable, forex or a salaried job?

And another question can be fed to the neural network as an output: is there a guarantee that trading successfully on Forex (suddenly) will last for a long time?

[Aleksei Kuznetsov]

Evgenii Shugurov #:
Ivan, input a question to the neural network: what is more profitable (profit/risk) and stable, forex or a salaried job?

And another question can be fed to the neural network as an output: is there a guarantee that trading successfully on Forex (suddenly) will last for a long time?

LLM will lie.
Statistics have calculated
https://www.mql5.com/ru/forum/86386/page3655#comment_55245431

PS. Although they say statistics lie too)))