Machine learning in trading: theory, models, practice and algo-trading - page 3699

New comment
 

To make it even simpler, he just took their probabilities instead of class labels to determine the "reliability" of predictions. That is, my code already implements partially the logic of conformal predictions, but without probabilities.

I'll compare the results now.

 

My code:

His code:

Too weak an improvement, often works worse than the original.

Again AI didn't make my life better :) although in the comments he keeps saying that the second approach is the most state-of-the-art and no better has been invented yet

But in terms of coding quality Gemini 2.5 is now definitely better than ChatGPT models and somewhere on Claude's level or higher. Probably this is the most correct and useful conclusion from everything that happened :)
 
Maxim Dmitrievsky #:

My code:

His code:

Too weak an improvement, often works worse than the original.

Again AI didn't make my life better :) although in the comments he keeps saying that the second approach is the most state-of-the-art and no better has been invented yet

But in terms of coding quality Gemini 2.5 is now definitely better than ChatGPT models and somewhere on the level of Claude or higher. Probably this is the most correct and useful conclusion from everything that happened :)

I use a Chinese - deepseek. Very often happy, but did not try to improve something of my own. Maybe you can try and post the result?

PS. By the way, Chinese is much superior to alice and chatgpt.

 
СанСаныч Фоменко #:

I use a Chinese guy - deepseek. I am very often happy with it, but I have not tried to improve something of my own. Can you try it and post the result?

PS. By the way, Chinese is much superior to alice and chatgpt.

Chinaman is somewhere on the level between chatgpt and Claude/Gemini. Well, or at the level of chatgpt. Sometimes I make him move his transistors. It is Claude and Gemini who are now showing more advanced answers in programming, in my opinion.
 
Do they understand, for example, why a virtual class is needed? How deep is the AI's own level of abstraction?
 
There is a strong book out
 
mytarmailS #:
There is a strong book out
Is it without codes, just theory?
Graph learning is interesting, haven't tried such methods yet.
 
Maxim Dmitrievsky #:
It's not coded, just theory?
Graph learning is interesting, haven't tried such methods yet.
No codes, more theory and formulas, no practical proof, and kinda wants to apply, with pictures... All in all 603 pages of water
 
Aleksander #:
No codes, more theories and formulas, no practical proof, and kinda wants to apply, with pictures... All in all, 603 pages of water
Not bad articles could turn out if you add practical examples. At least something related to trading, and not "this is it all".
 
Zy- in general water, not independent research, but a kind of analysis of the results of other works (researchers, with.....). Let's say, non-controversial conclusions.... Crossing out previous statements of their own.... In general, the water is wet, the sky is blue.
1...369236933694369536963697369836993700370137023703370437053706...3741
New comment