ForumSections

Machine learning in trading: theory, models, practice and algo-trading - page 2964

New comment
[Deleted]
mytarmailS #:
Exactly.

Not only that, I have a balance chart for 100 5min candles, that's less than a day, and this one has a balance for half a year

Question: where will the ideal balance look smoother?

Really, I'm writing this and I'm ashamed that I have to explain it.

And if there are very few signs, 3-5, will it be able to draw beautifully? The fewer signs, the better the test can be.
Maxim Dmitrievsky #:
And if there are very few signs, 3-5, will it be able to draw beautifully? The fewer signs, the better it can be on the test.
I think it depends on the data itself, if you take the data in the example, I think it can draw beautifully, but if you take real 100-dimensional data and shrink them by PCA to two-dimensional data, for example.

I don't know if it's 50/50.

It is better to include "good trading" on the test in the list of criteria in FF.
Maxim Dmitrievsky #:
And if there are very few signs, 3-5, will it be able to draw beautifully? The fewer traits, the better on a test can be.

That's a wonderful idea to create a teacher.

What would it look like in practice?

1. Create a set of traits for an unknown teacher.

2. We create a teacher based on the set of attributes.

3. Ask the question: will these traits have predictive power in the future for the created teacher?

4. We check the stability of the predictive ability in the future . If they do - lucky, but if they don't pass the test? Then how?


The problem is that the teacher is created for the traits, and the opposite should be the case.

[Deleted]
mytarmailS #:
I think it depends on the data itself, if you take the data in the example, I think he can draw beautifully, but if you take real 100-dimensional data and shrink them PCA, for example, to two-dimensional.

I don't know if it's 50/50.

It is better to include "good trading" on the TEST in the list of criteria in FF.
Well, it's worth digging. I'll soon finish my new creation, and I'll post what I've got.
[Deleted]
СанСаныч Фоменко #:

Wonderful idea for creating a teacher.

What would it look like in practice?

1. Create a set of attributes to an unknown teacher.

2. By the set of attributes we create a teacher.

3. Ask the question: Will these attributes have predictive power in the future for the created teacher?

4. We test the stability of the predictive ability in the future . If they will - lucky, but if they don't pass the test? Then how?


The problem is that the teacher is created for the traits, when it should be the other way round.

Why is that a problem? I've been doing it this way for a long time. There both attributes and teachers can be changed, like who forbids it :)
Maxim Dmitrievsky #:
Why is that a problem? I've been doing it this way for a long time. There you can search both signs and teachers, like who forbids it :)

I don't see any problem either, you can search for both features and targets in one algorithm....

but from the computational point of view it's very inefficient, but that's another issue.

About AI
[Deleted]
mytarmailS #:
About AI
h ttps://youtu.be/AW-4yBTf-XM

I didn't get into his reasoning, it takes a long time to express something.

but Katschyik was right about the big bang (according to the latest observations), about black holes it is not clear yet :)

Maxim Dmitrievsky #:

I didn't get into his reasoning, it takes him a long time to get his point across.

but Katschyik was right about the big bang (according to the latest observations), about black holes it is still unclear :))

Well there in the end he already about intelligence and Gpt says interesting things, but having missed the beginning, there is no sense to listen to the end.

Yeah, I watch Katschitzik regularly too
Who has tried to convert their models to ONNX and run in the tester?
1...295729582959296029612962296329642965296629672968296929702971...3749
New comment