Machine learning in trading: theory, models, practice and algo-trading - page 3702

 
Aleksander #:
Eggs are the grail of forex, all currencies go with eggs:

I've even got a special "egg beater" out of the back of my pocket:

the purple vertical line is the cursor, use it to move the pattern...other calculations.

in the picture, it crossed the red one, so it's a new egg, but upwards...

not ripe yet :-)


 
Aleksander #:
Eggs are the Grail of forex, all currencies walk in eggs:

I combined them with Elliott waves and got good predictions :).

They are embedded in my subconscious, through them I look at charts.
 
Maxim Kuznetsov #:

Any SB can be broken down into such self-similar sections, you will get a zigzag...on the sections our SB is slightly more "flatter" to 2 sigma (if it is counted from normal) than a completely stupid wandering.

Only this "a little" is not enough for stable filling of pockets :-)

The segments can be circled with an oval, and you can see the notorious "sine" inside (the kotir inside swings back and forth, slowly increasing amplitude and decreasing frequency).

This is all post facto, the effects of distributions and sensations.

They have self-similarity centres from which to reflect into the future. Post factum yes, but then you have to look at a larger TF and there you usually see a bigger chicken.

Elliott Waves describe just their different structures.

Then even econophysics was invented, I don't know what state it is in now.
 
Maxim Dmitrievsky #:
then we need to look at a larger TF.

On TFs below H4 the natural cycles of day, week, stock exchange opening/closing, money leaving/coming, from Asia to America and so on and so forth rule the roost and confuse everyone. And small events there sometimes look like a grand panic and a sharp turnaround, but in reality it's nothing.

but there you can scalp quickly (cycles and schedules are known) and trades go on for a foreseeable time without collecting a swap bag.

and at higher TFs you can already look at the beauty of theories and the egg-laying capacity of hens :-)

on the screenshot above "egg" was calculated as a solution "smart people with big money enter at the peaks and trade with maximum efficiency: outline the area of holding positions, outside the area of their positions there are no more positions. "

 
Maxim Kuznetsov #:

On TFs below H4 the natural cycles of day, week, stock exchange opening/closing, money leaving and coming, from Asia to America and so on and so forth, rule the roost and confuse everyone. And small events there sometimes look like a grand panic and a sharp turnaround, but in reality it's a flop

but there you can scalp quickly (cycles and schedules are known) and deals go on for a foreseeable time without collecting a bag of swaps.

and at higher TFs you can already look at the beauty of theories and the egg-laying capacity of chickens :-)

on the screenshot above "egg" was calculated as a solution "smart people with big money enter at the peaks and trade with maximum efficiency: outline the area of holding positions, outside the area of their positions there are no more positions. "

Couldn't find a good algorithmic way to describe fractals/multi fractals in terms of predicting them.

Somewhere in this thread I've uploaded an algo, an attempt at implementation. There was something very tricky. It is possible to estimate fractals by type of row entropy or dimensionality, but it is incomplete view. You need to add multi-windows, make affine rotations to compare pieces of graphs, adjust dimensions, etc. In short, the task is not for the faint-hearted.

All the transformations you do in your head in a few tens of seconds. Rotations, compressions, flips, comparisons with other TFs, search for similarities, evaluation of similarities, etc.

 
Maxim Dmitrievsky #:

GPT 4.1 offered an option, but cannot write working code. He can't fix the errors that occur either.

But he offered an interesting library for conformal predictions, which he tried to "put into the code". No credit so far, but thanks for the library, of course.

Researched

https://www.mql5.com/ru/articles/18324

Изучаем конформное прогнозирование финансовых временных рядов
Изучаем конформное прогнозирование финансовых временных рядов
  • www.mql5.com
В этой статье вы познакомитесь с конформными предсказаниями и библиотекой MAPIE, которая их реализует. Данный подход является одним из самых современных в машинном обучении и позволяет сосредоточиться на контроле рисков для уже существующих разнообразных моделей машинного обучения. Конформные предсказания, сами по себе, не являются способом поиска закономерностей в данных. Они лишь определяют степень уверенности существующих моделей в предсказании конкретных примеров и позволяют фильтровать надежные предсказания.
 

I made markup of deals based on self-similar structures, without details yet. And trained the MO on these markings.

It turns out to be quite stable on new data. I will tidy up the code and upload an article, the topic is interesting.

Partly proves the discussed theorem of eggs and chickens :)


 
Maxim Dmitrievsky #:

Researched.

https://www.mql5.com/ru/articles/18324

In this series of articles the Expert Advisor is position management based on received external signals.

Obtaining signals for the MO variant is ONNX, for the human variant is EX5-library compiled from mq5 code only.


Three definitions.

  1. Size in bytes of the human signal library - Human_Signal_Size;
  2. Size in bytes of ZIP-compressed signal MO-library - ML_Signal_Size.
  3. The size in bytes of the ZIP-compressed training sample - Data_Size.

double Koef1 = ML_Signal_Size / Human_Signal_Size;
double Koef2 = ML_Signal_Size / Data_Size;

What are the values of these two coefficients? And can a high value (greater than one) of the second coefficient speak about the probability of fitting for the MO (and for the variant invented by a human)?

 
fxsaber #:

In this series of articles, the Expert Advisor represents the management of positions based on the received external signals.

Receiving signals for the MO variant is ONNX, for the human variant - EX5-library compiled only from mq5 code.


Three definitions.

  1. Size in bytes of human signal library - Human_Signal_Size;
  2. Size in bytes of ZIP-compressed signal MO library - ML_Signal_Size.
  3. Size in bytes of ZIP-compressed training sample - Data_Size.

What are the values of these two coefficients? And can a high value (greater than one) of the second coefficient speak about the probability of fitting for the MO (and for the variant invented by a human)?

It's hard to say in this proportion, models are different. Usually the number of model parameters (number of features) is compared to the size of the dataset (number of training examples). The dataset should be at least 2-3 times larger.

For example, an elephant can be described by three features or 50000000000 features. Which model is more fitted? There is no answer because no fit criterion is introduced. Let's say that based on the appearance of the stuffed elephant both models will fail to predict the elephant.
 
Maxim Dmitrievsky #:

They do have self-similarity centres from which to reflect into the future. Post facto yes, but then you have to look at a larger TF and there you usually see a bigger chicken.

Elliott Waves describe just their different structures.

Then even econophysics was invented, I don't know what state it is in now.
I figured out how to study such structures by means of MO. I've written a few approaches, but there are still options. As a replacement for the old efficient market hypothesis, the multifractal hypothesis looks more interesting. There are special packages to test the hypothesis. Re-reading Mandelbrot even for this to refresh.