Machine learning in trading: theory, models, practice and algo-trading - page 3129

New comment
 
Aleksey Vyazmikin #:

I don't think it's your place to give me advice.

There is such a park in youtube, when first they ask a passerby how to go to McDonald's, and then ask him not to tell them what to do.

A trained model with and without the trait is compared. The difference is defined as the ATE. The deck must be shuffled to remove the score bias. Sounds like some kind of mantra if you don't read the theory.
 
Maxim Dmitrievsky #:
There is such a parknk in youtube, when first they ask a passerby how to go to McDonald's, and then ask them not to tell them what to do.

A trained model with and without the trait is compared. The difference is defined as the ATE. The deck must be shuffled to remove the score bias. Sounds like some kind of mantra if you don't read the theory.

That's the point - you're talking about things that everyone has known and prescribed for a long time. I'm saying why it's not optimal and why it doesn't work, but you don't hear or understand it.

 
Aleksey Vyazmikin #:

That's the thing - you are talking about things that are known and prescribed for a long time. I say why it is not optimal and why it does not work, but you do not hear and do not understand it.

It works and is optimal, the theorem is proven, which has already been mentioned. + articles.

But it's not for me to tell you
 
Maxim Dmitrievsky #:
Works and is optimal, the theorem is proved, which has already been mentioned. + articles.

But it is not for me to tell you

And you don't need to point - you can just have a sane dialogue, and argue your position.

I wrote that such methods are based on the use of a model, so all measurements are relevant for the model, not for a single predictor - this is the disadvantage.

If you have any result showing the effectiveness of the proposed method, please let me know - it would be interesting and constructive for discussion.

 

I highly recommend you to reconsider the very approach to feature construction. Functional transformations don't work here, the market is NOT a time series!

Here is the M1 entry point.


here is the reason for entry and the H1 entry price.

=========================================

And all your signs do is to look for patterns/curves in the last 5-10 candles.


It's much more complicated than that.

 
Aleksey Vyazmikin #:

And you don't need to point it out - you can just have a sane dialogue, and argue your position.

I wrote that such methods are based on the use of a model, so all measurements are relevant for the model, not for a single predictor - that's the disadvantage.

If you have any result showing the effectiveness of the proposed method, please let me know - it would be interesting and constructive for discussion.

What is a single predictor evaluated through?

If there is a predictor, then it has a function and a target function? )
 
mytarmailS #:

I highly recommend you to reconsider the very approach to feature construction. Functional transformations do not work here, the market is NOT a time series!

Here is the M1 entry point


here is the entry reason and entry price H1

=========================================

And all your signs are doing is looking for patterns/curves in the last 5-10 candles.


It's more complicated than that.

It's just fractal logic.
 
spiderman8811 #:
The usual fractal logic.

Can you show me something like that?

If it's that common.

 
mytarmailS #:

I highly recommend you to reconsider the very approach to feature construction. Functional transformations do not work here, the market is NOT a time series!

Here is the M1 entry point


here is the entry reason and entry price H1

=========================================

And all your signs are doing is looking for patterns/curves in the last 5-10 candles.


It's more complicated than that.

How do you know what anyone's doing? That's a bit of a boast. I have signs that take into account movements over the last three months.

 
Maxim Dmitrievsky #:

What is a single predicator evaluated through?

if there is a predictor, then it has a function and a target? )

Of course, it is obvious - there is a target and some logic linking them.

Earlier in a separate thread I showed that it is possible to build different models from the same data - I get about 30% profitable and 70% unprofitable. That is why I think that the method of estimation through the model is not very reliable.

1...312231233124312531263127312831293130313131323133313431353136...3659
New comment