Machine learning in trading: theory, models, practice and algo-trading - page 1698

 
Aleksey Vyazmikin:

Go ahead and post it.

Doc was looking for another way to select predictors - by building a decision tree on genetics - we tested it together before he left.

This is a more serious approach. Vtreat is just a pre-processing of data for further final selection of inputs during optimization. That's how I do it.

Attached is the script for R, and I also attach my file. Funny, run it and see how much it can reduce the raw data. There in the script you need to specify the path to the file to read and the path where to write the result.

Files:
111.zip  2961 kb
 
Tag Konow:
Well... I was beginning to penetrate into the essence of MO from your posts, and here is such a confession ...))

Although, I wonder, perhaps I will understand MO from the point of view of a drunken man and create the first drunken NS in the world?)))
You can't read the experience. The main thing in MO is to understand the philosophy. It is very important to answer the main question of what the network can and can not. Not infrequently, it is the inflated expectations of beginners that lead them to failure and, consequently, to disappointment in the tool as a whole. Do not expect anything supernatural from it and you will not be too disappointed. But you won't find a better tool as an assistant.
 
Mihail Marchukajtes:
You can't sell your experience. The main thing in MO is to understand the philosophy. It is very important to answer the main question of what networks can and cannot do. It's not uncommon that inflated expectations of beginners lead them to failure and, as a consequence, to disappointment in the tool as a whole. Do not expect anything supernatural from it and you will not be too disappointed. But you won't find a better tool as an assistant.
So, the main question is unclear. What can networks do and what can't? The human brain also has a neural network and it can do EVERYTHING. Artificial NS are limited, but it is not clear what. The number of neurons? The training sample? Imperfection? It's not clear...
 
Tag Konow:
So, the main question is unclear. What can networks do and what can't? The human brain also has a neural network, and it can do EVERYTHING. Artificial NS are limited, but it's not clear what. The number of neurons? The training sample? Imperfection? It's not clear...
Well as an example of philosophy. NS has a computing capacity and the task of the engineer to reduce the load on the NS as much as possible pre-processing, etc., and not to shove her whatever you want, like she will figure it out. No it won't.
 
Tag Konow:
So, the main question is unclear. What can networks do and what can't? The human brain also has a neural network and it can do EVERYTHING. Artificial NS are limited, but it's not clear what they are. The number of neurons? The training sample? Imperfection? It's not clear...

Restricted by Teacher !!!!!!

 
mytarmailS:

Limited to the teacher !!!!!!

The teacher supplies data to the input of the NS for learning. That is, one particular NS is designed to solve one, strictly specific task. If we combine a set of NS and teach each NS to solve its task, then we get an analogue of a brain? That is, a machine solving many problems. Or, is that not enough?
 
Tag Konow:
Does the teacher supply data to the input of the NS for learning? That is, one particular NS is designed to solve a strictly specific problem. If we combine a lot of NS and teach each NS to solve its own problem, we get an analogue of a brain? Or is it not enough?

On the face of a clear overexpectation of the tool. Believe me, even the answer to a specific task in the form of "Yes", "No" will already be enough and unnecessary complication will be unnecessary.

Just learn to see the benefit in a small and believe me the result of this benefit can be tremendous, because you are armed with this tool and your opponents are not...

 
Konow Retag:
The teacher supplies data to the input of the NS for learning. That is, one particular NS is designed to solve one, strictly specific task. If we combine a lot of NS and teach each NS to solve its task, we get an analogue of a brain? That is, a machine solving many problems. Or, is that not enough?

no! it is not enough.

A neural network or other AMO is just a "multidimensional optimization" and that's it!

It is a tool to solve the problem, that's all!

And the problem has to be set!

And the problem has to be invented!

And the task must be selected from other tasks!

And the task is relevant!

This is all up to the man so far... They call this kind of creativity...

 
Mihail Marchukajtes:
There is a clear overexpectation of the tool. Trust me, even a "Yes" or "No" answer to a specific question will be enough and unnecessary complications will be unnecessary.
I do not understand what is the problem with creating an NS that recognizes price patterns? A human can do it right away, without training. And they teach and teach... Where is the NS that recognizes basic chart patterns? Why isn't there one in QB? There are a lot of algorithms, but there is no such network...

There are networks that recognize faces, numbers, road signs, cardio rhythms, and even emotions. And what about price patterns?
 
Tag Konow:
I don't understand - what is the problem with creating a NS which recognizes price patterns? The man can do it at once, without training. And they teach and teach ... Where is the NS that recognizes basic chart patterns? Why isn't there one in the QB? There are a lot of algorithms, but there is no such network...

There is no problem creating such a network , and it will also recognize patterns no worse than a human, even better...

The funny thing is that the network will not make money on these patterns just as well as a person who recognizes them at once )) !!

the problem is with the patterns, not the network. and the teacher who wants the network doesn't know what he wants, but he thinks he does.