Machine learning in trading: theory, models, practice and algo-trading - page 3399
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Vorontsov talks in the video, you've seen it. About the concept of fundamental models, starting at one o'clock.
Oh, I remember.
So it's kind of a concept, any information can be reduced to a vector, he's talking about eembeddings, well, yes it is.
But it's just a concept, and LLM is trained on embeddings of text and nothing more, its whole structure is trained for it.
So, if you decide to give it your own OHLC embeddings, nothing will happen )))
You need to train it from scratch on different embeddings, different tasks, different targets ... at the same time, so that it writes text and draws and speaks, etc... multimodality, in other words.
And you want to take a neura who can text and give her OHLC )) it won't work.
Ah, look, I remember.
So it's kind of a concept, any information can be reduced to a vector, he's talking about eembedding, yeah, that's right.
But it's just a concept, and LLM is trained on text embeddings and nothing more, its whole structure is trained for it.
So, if you decide to give it your own eembeddings with OHLC, nothing will happen )))
Knowledge distillation may happen, when the model transfers its general "knowledge" to your sphere. I'm a little fuzzy on it, but it goes something like this.
It can translate into numbers and text and pictures and even sound, and into programme code.Knowledge distillation can occur when the model transfers its general "knowledge" to your domain. I'm a bit fuzzy on this, but it's something like this.
Nothing's gonna happen.
Imagine a PCA trained on some data, that's the guts of a neuron.
You add unfamiliar data to it, it's just a sort of isolated cluster of points that Neura hasn't met and has no idea what to do with it.
She won't transfer the knowledge because that cluster is in coordinates she's never worked in before.
Nothing's gonna happen.
Imagine the PCA trained on some data, that's the guts of Neira.
If you add unfamiliar data to it, it will just appear as a kind of isolated cluster of points that Neura has never met and has no idea what to do with it.
She won't transfer the knowledge because that cluster is in coordinates she's never worked in before.
Well, that's what everybody does nowadays, they train them for their tasks. I'll have to look for examples.
You will update the weights (links) of the model, not add new points.Well, that's what everybody does nowadays, they train them for their own tasks. I'll have to look for examples.
You're updating the weights of the model, not adding new points.You don't get it. I'll leave you to it.
You don't get it. I'll leave you to it.
I get it, but that's not how it works. It encodes any information into semantic vectors, it doesn't matter what nature the data is, it's just symbols. It already knows all those symbols, it's the sequence that matters.
This one is good too https://youtu.be/EutX1Knl6v4?si=bBicqPoh474EiRIc
Yes, it's a good one, heard a lot of things before, but there are some new ideas for me. And, funny enough, I found a similar algorithm for something I have planned in development, with a similar goal.
Probably there are packages ready for R, have you tried to do something like rating the result of selection by different methods?
I implemented approaches in MQL in semi-manual mode, but it was troublesome. Maybe later I will do it in Python.
Yes, good one, heard a lot of things before, but there are some new ideas for me. And, funny enough, found a similar algorithm for something I have planned in development, with a similar goal.
I guess there are packages ready for R for all this, have you tried to do something like rating the result of selection by different methods?
I implemented approaches in MQL in semi-manual mode, but it was troublesome. Maybe later I will do it in Python.
What do you mean by that?
Like to find out which method of feature selection is better?
Basically there are only two types, exhaustive search and euristic search methods (discrete optimisation).
Full search is always better, but not always possible if there are a lot of features. In addition, we are not looking for one best feature, but the best subset of princes, on more or less normal dataset to do a complete search is impossible because of combinatorial explosion, so euristic - discrete optimisation (but without guarantee that found the best solution).
There is one good package, I haven't tested it deeply and I don't know the mathematics, but guys claim that the algorithm finds the best subgroup in polynomial time (very fast), i.e. it's not a complete search and not euristics. I used it a little bit, I think the package does what they say. So basically this is the leader of the selection methods.
https://abess-team.github.io/abess/
I think there's one for python, too.
====================================
And the point (the essence) is not even in the selection of prizes (although it is necessary), but in the generation of features of applicants. This is the most important thing, this is the most important thing, these are sensors, eyes and ears of the trading system.
The article Working with ONNX models in float16 and float8 formats has been published:
Форматы данных, используемые для представления моделей машинного обучения, играют ключевую роль в их эффективности. В последние годы появилось несколько новых типов данных, разработанных специально для работы с моделями глубокого обучения. В данной статье мы обратим внимание на два новых формата данных, которые стали широко применяться в современных моделях.
In this article, we focus on two such new data formats, float16 and float8, which are beginning to be actively used in modern ONNX models. These formats are alternatives to the more accurate but resource-intensive floating-point data formats. They provide an optimal combination of performance and accuracy, which makes them particularly attractive for various machine learning tasks. We will explore the main features and advantages of float16 and float8 formats, and present functions to convert them to standard float and double.
This will help developers and researchers to better understand how to effectively use these formats in their projects and models. As an example, we will consider the work of the ONNX model ESRGAN, which is used to improve the quality of images.
Author: MetaQuotes