Machine learning in trading: theory, models, practice and algo-trading - page 3711

[Aleksey Nikolayev]

Maxim Dmitrievsky #:
Another article bot went to the top of the market. Breaking records for the demand for material :) While promptus are fighting for the place of being the most tedious on the forum. I counted 5-7 people who have made good money in the moment, on the basis of articles.

The number of readings of articles, compared to others, sometimes in times different.

Imho, the main thing in this business - to have an established structure for the sale of products. Fresh ideas are of course also important - without novelties, you can fall out of favour.

[[Deleted]]

Aleksey Nikolayev #:
In my opinion, the main thing in this business is to have a well-established structure for selling products. Fresh ideas are of course also important - without new products you can fall out of favour.

Well, these are not essential products, so to establish a structure is not always possible :) even stalls with beer often go bankrupt.

[[Deleted]]

Almost completed the algo on the mix of experts. It turns out to be quite interesting. It remains to rewrite the tester for this case. Router-model is often trained with acceptable error, it will be interesting to see in the tester and on new data :).

Another important part is the selection of expert models according to different criteria. A large selection of possible actions.

Example of selection of experts by thresholds and minimum number of correctly predicted labels. Each expert is trained on its own data.

--- Результаты распределения меток ---
Модели, которые были использованы и соответствуют порогу min_assigned_samples:
  Модель 1: 1425 назначенных примеров
  Модель 5: 4057 назначенных примеров
  Модель 9: 893 назначенных примеров
  Модель 11: 9063 назначенных примеров
  Модель 14: 1090 назначенных примеров
  Модель 15: 747 назначенных примеров
  Модель 16: 185 назначенных примеров
  Модель 17: 212 назначенных примеров

And then a router-model that selects which expert is trading at the moment

>>>  router_model.get_best_score()
{'learn': {'Accuracy': 0.8489833481127215, 'MultiClass': 0.4767885754992078}, 'validation': {'Accuracy': 0.776081408038278, 'MultiClass': 0.6515423154292276}}

[[Deleted]]

Well... at least it works without errors :) otherwise - expectedly

[[Deleted]]

The same bias-dispersion problem arises as with conventional learning. Fewer experts - lower variance, more experts - lower bias. More experts - higher learning error of the router and vice versa.

[[Deleted]]

Everything, the latest LLMs in Pieces.app have become paid. Even Gemini 2.5 flash, which is basically free. Thanks to your house, deleting this rubbish :)

[[Deleted]]

Two articles on fractal analysis and ML, theory and practice:

https://www.mql5.com/en/articles/18453

https://www.mql5.com/en/articles/18566

[Aleksey Nikolayev]

Maxim Dmitrievsky #:

The whole point of "AI" in one screenshot. Provides links to my articles.

You can try asking in English. It usually works for Perplexity.

Imho, there are few links to advanced topics in Russian in general.

[[Deleted]]

Aleksey Nikolayev #:

You could try asking in English. It usually works for Perplexity.

Imho, there are few references to advanced topics in Russian in general.

Where did the auto-translation from English sources go?) Not for myself, I just wanted to see what he would find and find. It turned out to be even more trivial. Well and some proof of the uniqueness of the material.

[[Deleted]]

Dipsic one of the links hinted at the possibility of selling books electronically on ozone, for example on MO in trading. You can use this option to achieve great fame and power :)

One can become more famous than Prado and Pugacheva.

A startup option with zero investment and the possibility of monetising LLMs as scribblers of pieces of text.

Then create an investment fund in dubais and a consulting agency to attract big money.