Machine learning in trading: theory, models, practice and algo-trading - page 3422

New comment
 
Aleksey Vyazmikin #:

Where to look for them, except for your code? I see in the library code that this is how data is read - by cell address - I did it by analogy.

We get

Have you tried it, does it work?

Forum on trading, automated trading systems and testing trading strategies.

Errors, bugs, questions

Aleksey Vyazmikin, 2024.03.12 05:51 pm

Oh, how it happens - overnight they corrected the syntax now the value should be assigned in this way.

MatrixLearn.Set(row,col,Input_arr_Data[N_Stolb*row+col]);

The old version is no longer compiled - you can get schizophrenia that way...


Now I started counting clusters!

 
Maxim Dmitrievsky #:

There is another non-obvious thing that can affect the training results. It is, for example, to train a classifier not only to predict buy/sell labels, but at the same time to train it to classify SEALs (rough example). That is, teach the main task and various side tasks.

This might have something to do with the internal structure of the model. I haven't seen any such studies.

Made these kind of SEALs.

What other SEALs might be suitable?


 
What if we imagine that the chart is not random, but there are just too many patterns. And we have very little history yet, so we can't compare patterns. Therefore, history does not repeat itself for us. Then we definitely can't do without augmentation methods. But generating history from the past of your own series is a way to nowhere.
 
Aleksey Vyazmikin #:

Now started counting clusters!

Clusters
 
Maxim Dmitrievsky #:
And if we imagine that the chart is not random, but just too many patterns. And we have very little history yet, so we can't compare patterns. Therefore, history does not repeat itself for us. Then we definitely can't do without augmentation methods. But generating history from the past of your own series is a way to nowhere.
It's easy to check this if you take one single pattern and calculate the statistics. And it will be seen that there are patterns and they are repeated, but the statistics on them is not repeated.

The market rather consists of an aditive structure those patterns from different components are superimposed on each other, and their sum is the price. If you look at the sum, almost nothing ever repeats if you look at the individual components, it is already better. Here will help decomposition rsa, fourier, etc.
 
Aleksey Vyazmikin #:
And this drawdown you're having, is it within the framework of the ts or what went wrong?
 
Aleksey Vyazmikin #:

Now started counting clusters!

I tried it about 4 years ago. Looks like they changed the syntax in the meantime. Glad you were able to figure it out.
 
Maxim Dmitrievsky #:
clusters

Will you try to prove it? ;)

Maxim Dmitrievsky #:
And if we imagine that the chart is not random, but just too many patterns. And we have very little history yet, so we cannot compare patterns. Therefore, history does not repeat for us. Then we definitely can't do without augmentation methods. But generating history from the past of your own series is a way to nowhere.

That's the way it is in essence - there are patterns with low response in the leaves.

 
mytarmailS #:
And this drawdown you have, is it within the framework of the TS or what went wrong?

Within the framework of the TS, everything is worked out according to the settings.

 
Forester #:
I tried it about four years ago. Looks like they changed the syntax in the meantime. Glad you were able to figure it out.

It was changed overnight - that's what I wrote...

1...341534163417341834193420342134223423342434253426342734283429...3691
New comment