Machine learning in trading: theory, models, practice and algo-trading - page 3534

 
Maxim Dmitrievsky #:

Will need to be more immersed in the discipline of statistical learning to lead the discussion

If you find something new - write me - I will listen with interest!

 
Aleksey Vyazmikin #:

If you find something new - write - I'll listen with interest!

It's not a priority at the moment. I just don't like the very approach of picking at data when there are universal solutions. They may not be as accurate and transparent, but they get rid of routine.

I realise that I can no longer explain my approach in normal human language either. Pure empiricism, without theory.

That is, first you just have to do something, let's say, for a year, and see that it is good. And then spend another 10 years trying to give a theoretical justification for why it works. :)

 
Maxim Dmitrievsky #:

It's not a priority right now. I just don't like the approach of picking at data when there are universal solutions. They may not be as accurate and transparent, but they get rid of the routine.

In this company I am looking for assurance that I will not lose money by investing it once in a TS. It seems that the better you understand and control the process, the more confident and less anxious you are.

I am ready to discuss any approach, so tell me about yours :)

Maxim Dmitrievsky #:

I realised that I can't explain my approach in normal human language. Pure empiricism, without theory.

So, let's see, and we will come to mutual understanding ...

Maxim Dmitrievsky #:

That is, first you just have to do something, let's say, for a year, and see that it is good. And then for another 10 years try to give a theoretical justification for why it works. :)

I agree, that's why I still want to try my method on some data with an existing pattern. I haven't found anything suitable in the public domain yet.

 

Also, I found it interesting that the Gini index, which was used to select the quantum cutoff for the drop, grows from iteration to iteration

And here is the probability shift in this predictor on the delayed sample for the zero class within the selected quantum segment.

 

Some more information to ponder.

Here is the value of balance (in points with a step of 0.00001) at each iteration on the sample train when selecting a quantum segment by the Gini index

And the change in the probability of choosing the class "1" (the average of 0 and 1 in the sample) - also from iteration to iteration (I took the last variant - there is a percentage and bigger) - the graph below.

And, let's take a metric not from classical MO, but purely on money - it's called LossFactor :).

Same two graphs.


The difference in probability growth (second graph) is 15%, but the situation is the opposite for the balance. So you should think what you need in the end - more profitable trades or more profits in a trade....

In general, this is the difference between the terminal optimiser with standard metrics and MO methods.

You need some kind of more balanced measuring device.

 
I would also talk about quantum entanglement theory, but I'm afraid we'll fall into terminology discussion again :))))))
 
Aleksey Vyazmikin #:

In this company I am looking for confidence that I will not lose money by investing it once in TC. It seems that the better one understands and controls the process, the more confidence and lower anxiety.

I am willing to discuss any approach, so tell me about yours :)

In this way, we may come to mutual understanding....

I agree, that's why I still want to try my method on some data with an existing pattern. I have not found anything suitable in the public domain yet.

I have everything, I need sometimes testers in different DCs and sometimes investors. Because I am often too lazy to even open an account myself, I am engaged in high creativity, not some household stuff ☕️.

By the way, "investor" started to be interested in "AI" topic in trading. By investor I mean not someone specific, but a collective image of cheliks.

And a lot of cheliks who write algos for their investors, refocus on MO at least.

 
Maxim Dmitrievsky #:
I have everything, I need sometimes testers in different DCs and sometimes investors. Because I am often too lazy to even open an account myself, I am engaged in high creativity, not some mundane stuff ☕️

By the way, "investor" started to be interested in "AI" topic in trading. By investor I mean not someone in particular, but a collective image of people.

And a lot of cheliks who write algos for their investors, reorient to MO at least.

h ttps://youtube. com/shorts/dB0oLlMMMUI?si=a7KTFDv42JGlhs4f

I don't have a queue of investors to rely on other people's money, and other people's money is more responsible than my own.

 

I'm fooling around with an invented method of marking trendlines without a teacher, based on clustering.

It's pretty interesting


 
mytarmailS #:

I'm fooling around with a method I invented for marking trend lines without a teacher, based on clustering.

It's pretty interesting


"whatever the child does, as long as it is not with his hands" :-))