Machine learning in trading: theory, models, practice and algo-trading - page 3576

New comment
 
Label smoothing/relaxation - interesting regularisation techniques. In the piggy bank of the fact that data partitioning is strongly important.

 
СанСаныч Фоменко #:

Not exactly.

If we talk about the valuation of the classification, there are no earnings estimates there. There are their own estimates and their own packages (ROC).

I have been training regression models rather than classification models for the last few years. Just the financial result from the transaction and I submit it as a teacher.

 
Forester #:

I have been teaching regression models rather than classification models for the last few years. Just the financial result from the transaction and submit it as a teacher.

And what is the result?
 

Ostolopia has gone off the scale again. time to ask LLM :))))))

In the context of machine learning with a teacher, the teacher provides not only labels but also features (features). Let's look at this in more detail:

  1. Teacher(Supervisor):

    • Role: As already discussed, the teacher provides the training data, which includes both features and labels. This can be a person who collects and labels the data, or an automated system that already contains this information.

    • Function: The teacher provides the model with all the necessary information for learning, including the characteristics (features) of the objects and the correct answers (labels) for each object.

  2. Features:

    • Role: Features are characteristics or properties that describe each object in the training dataset. They are inputs to the model.

    • Function: Features allow the model to distinguish between objects and make predictions based on these differences. For example, in an image recognition task, features can be image pixels or more complex features extracted from an image.

  3. Labels:

    • Role: Labels are the outputs that the model is trying to predict. They indicate the correct class or value for each feature.

    • Function: Labels serve as a target for training the model. The model uses features to predict labels and compares its predictions with the actual labels to adjust its behaviour and improve accuracy.

Relationship between teacher, features, and labels:

  • The teacher provides data consisting of traits and labels.

  • Traits are the input data that the model uses to make predictions.

  • Labels are the output data that the model tries to predict and that guide learning.

Thus, the teacher provides not only labels, but also features, which together allow the model to learn and make accurate predictions.

 
mytarmailS #:
What's the result?

Same as everyone else. 50/50.

Or if there are profitable options, then the drawdowns for a year - two. That's not good either.

 
Maxim Dmitrievsky #:
Role: As already discussed, the teacher provides the instructional data, which includes both attributes and labels. It could be a person who collects and labels the data, or it could be an automated system that already contains this information.

Here we go. You say the teacher is only a human. With our amount of data up to millions of rows a year. It's silly to hire a human and label every single one of them. It's all programmatically solved.

 
Forester #:

Here we go. You're saying that a teacher is only human. With our amount of data up to millions of rows a year. It's silly to hire a human to mark up every single one of them. It's all done in software.

I'm not confusing teacher and tagging. And I didn't say it had to be human

Just like with AI, it's difficult to translate. It's English for supervised learning. You don't confuse tagging with control. It's a different concept.

Then you could call it "tagged learning", but it's not called that, so the meaning has to be different.
 
Forester #:

Like everyone else. 50/50.

Or if there are profitable options, then drawdowns for a year or two. That's not good either.

Make a minimal clustering into several states and try to learn and trade separately on each one.
After different attempts, the results may improve. They can improve a lot if you do it sensibly, like in Alexey's article.
 
Maxim Dmitrievsky #:
Do minimal clustering into multiple states and try to learn and trade separately on each.
I will postpone it for the future. I am busy with other things, but I will soon finish and start working on MO again.
 
Maxim Dmitrievsky #: like in Alexei's article.

can I have the link?

1...356935703571357235733574357535763577357835793580358135823583...3656
New comment