Machine learning in trading: theory, models, practice and algo-trading - page 227
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
Well, the point can be summed up in a set of sentences. That's all I ask.
Now the task is not to study the topic, but to preliminarily estimate its scope. That's why I say formulate it (if you understand).
See at least the first lecture. You can't do it in a couple of sentences, it's too vast an area.
You see, I'm waiting for you to articulate the point, because I want to understand exactly what you want by the concept of machine learning.
What is told in the lecture is the understanding of other people. Perhaps the algotrading community needs specific machine learning.
I want to understand what exactly algotraders need in the broad realm of machine learning and thus reduce time spent learning irrelevant areas, limit code bypassing unnecessary tasks, and ultimately achieve the right implementation of the goal.
You see, I'm waiting for you to articulate the point, because I want to understand exactly what you want in terms of machine learning.
What the lecture will tell you is other people's understanding. Perhaps the algotrader community needs specific machine learning.
I want to understand what specifically algotraders need in the broad realm of machine learning and thereby reduce time spent learning irrelevant areas, limit code bypassing unnecessary tasks, and ultimately achieve the right implementation of the goal.
There are two tasks here, in very general terms.
1) qualitative feature selection
The features are :
For example, you like technical analysis, supports, resistances, rebounds, breakdowns, etc..
...you see the market through these features, so we do not show prices but signs of supports, resistances, rebounds and breakdowns... into the algorithm
and here comes the point
2) decision generation
the algorithm "juggling" these signs starts to create some optimal trading rules - decisions, and it selects those signs that are worth something and those that are not important for making a good decision
=====================
so proper data processing is 98% of the work
The training of the MO is 2%.
There are two tasks here, if very generally
1) qualitative selection of features
These are signs:
For example, you like technical analysis, support, resistance, rebound, breakdown etc..
You see the market through these features, so we do not give the prices but only those features that support, resistances , rebounds, breakdowns etc. ... into the algorithm
and here comes the point
2) decision generation
The algorithm "juggling" with these signs starts to create some optimal trading rules - solutions and selects those signs that are worth something and those that are not important for making a good decision.
Here, thank you. I'm starting to get a picture.
It's a kind of general collection and analysis of data change signatures in the current period, which are fed into a special algorithm, analyzed there, statistical data signatures are collected, patterns and repeats of signatures are investigated and decisions about system behavior are generated.
About this?
You see, I am waiting for you to formulate the essence, because I want to understand what exactly you need under the concept of "machine learning".
What will be told at the lecture is other people's understanding. Perhaps the algotrading community needs specific machine learning.
I want to understand what specifically algotraders need in the broad realm of machine learning and thereby reduce time spent learning irrelevant areas, limit code bypassing unnecessary tasks, and ultimately achieve the right implementation of the goal.
The essence of MO is approximation, to get a quasi-model of dataset that generates it. In the case of classification it is a cloud of marked points to get masks separating them.
The essence of MO is approximation, to get a quasi-model of dataset that generates it. In the case of classification it is to get masks separating them from the cloud of marked points.
Approximation is a generalization of values. That is, the conclusion of different data values inside a selected range? Further, you can create a numerical model that generalizes the change of some value over a period of time. By collecting these models, you can create statistics to guide your decisions and choices of actions.
Am I going in the right direction?
In short, -
1. Create an algorithm that collects streams of values of any parameters (data) we need and runs them through the ring buffer.
2. We pass the streams of values stored in the ring buffer through a special filter, which generalizes them to the ranges of these values.
3. A generalized (with the help of the ranges) digital model of the nature of variation of each parameter value in the ring buffer is created, and written in the appropriate format.
4. This model is sent to the statistical algorithm collecting these models.
5. We loop through the database containing models (signatures) of the nature of changes of values of our parameters and find the model that best suits the current situation.
6. A decision is made about how the system will behave in the situation captured in this signature (model).
I will formulate it more precisely later.
Whattoxic has shown is a kind of clustering, but with a teacher, the points in the beginning are signs or rather their numerical parameters, you have a target buy and sell, so before the training you marked out where there was growth (buy) and decline (sell), and the algorithm begins to stupidly divide parameters of signs by target, like the blue area is buy, red is sell...
But right now the latest flavor is something like this
https://www.youtube.com/watch?v=05rEefXlmhI
https://www.youtube.com/watch?v=qv6UVOQ0F44
https://www.youtube.com/watch?v=xcIBoPuNIiw
but I am a total nerd.
This one is hilarious)))
https://www.youtube.com/watch?v=pgaEE27nsQw
Whattoxic has shown is a kind of clustering, but with a teacher, the points in the beginning are signs or rather their numerical parameters, you have a target buy and sell, so before the training you marked out where there was growth (buy) and decline (sell), and the algorithm begins to stupidly divide parameters of signs by target, like the blue area is buy, red is sell...
But right now the latest buzz is something like this
https://www.youtube.com/watch?v=05rEefXlmhI
https://www.youtube.com/watch?v=qv6UVOQ0F44
https://www.youtube.com/watch?v=xcIBoPuNIiw
but I'm a total nerd.
I'll take a look at it all tomorrow.
Two years ago I had ideas, as it turns out, similar in some ways to machine learning. I called it "collecting digital signatures of parameter value changes". I thought up a basis for this technology and wrote it down. I never got around to implementing it, because I was always distracted by other things.
Tomorrow I will describe the whole concept of these "signatures" and you tell me how akin it is to machine learning.
If they are close things, then the technology of creating algorithms is already clear to me.