NS + indicators. Experiment. - page 7

New comment
 
Prival:
I agree that it is promising and interesting. But you are wrong about the fact that it is impossible to logically understand "why a neural network made exactly such a decision". The logic of decision making can and must be programmed (this is how it is done in normal algorithms).

I do not agree. A neural network is a mini-brain. Only a human has a few hundred thousand neurons, while a neural network has a few units or tens. How can a brain be logically predicted? It is a kind of "prediction prediction". You can't control a neural network's decision making. That's what a neural network is..... You can only make or not make that decision. But that's another topic.....
 
LeoV:
Prival:
That it is promising and interesting I agree with that. But you are wrong about the fact that it is impossible to logically understand "why a neural network made exactly such a decision". Decision logic can and must be programmed (this is how it is done in normal algorithms).

I do not agree. A neural network is a mini-brain. Only a human has a few hundred thousand neurons, while a neural network has a few units or tens. How can a brain be logically predicted? It is a kind of "prediction prediction". You can't control a neural network's decision making. That's what a neural network is..... You can only make or not make that decision. But that's another topic.....
Perhaps I have not accurately explained. How NS works and how it makes decisions is also known. And the algorithm is not a brain yet, it is far from it. It is interesting what you would say about military targeting algorithm, which does not know how it decides what target to aim at (at Moscow or New York :-)). But there is a lot of what I said, it is also a NS in your understanding.
 
LeoV:
So the NS is essentially a black box. In the sense of making a decision. It is impossible to logically understand why a neural network made that particular decision. That's why it's a neural network.

You seem to make some alien miracle out of NS. There is a quite developed mathematical apparatus, and in general it is known what NS does in the input space. And knowing it we may understand why the net took this or that decision. And the comparison with a brain... It is not only a question of network capacity (the number of neurons) but the brain is more complicated. That is what numerous recent studies indicate. It is by no means a black box, otherwise how would you work with it? Then it would remind Strugatsky's "Picnic by the Side of the Road" - a lot of things not clear how they are made, not clear how they work, it is not clear what they were originally designed for and from some learned to extract benefit. And I agree about perspective, it is there, thanks to the fact that this apparatus is still underdeveloped, which means that more than one small miracle can still happen.
 
Prival: I may not have explained it accurately. How the NS works is known and how it makes decisions is also known. And the algorithm is not a brain yet, it is a long way from that. It is interesting what you would say about combat targeting algorithm, which does not know how it decides which target to aim at (Moscow or New York :-)). But there is a lot of what I said, there is also NS in your understanding.

Well, I don't really give a damn - count it as you like. And I didn't say how, but why.... Write a "missile guidance system" in MQL - I'll be happy for you.....
 
TedBeer:
And knowing this you can understand why the network made a particular decision.
If you understand why the neural network made that decision, then use that knowledge - you don't need a neural network to trade......
 
LeoV:
If it's clear why the neural network made that decision, then use that knowledge - you don't need a neural network to trade......
I do, because I'm lazy and don't want to trade with pens :-)
 
TedBeer: Well you do, because I'm lazy and don't want to trade pens :-)
Well with neuronets you also need to work, not be lazy..... Or are you one of those people who are looking for the grail and want to go to the island and rest, while a neural network will work?
 
LeoV:
Well, neural networks also need to work, not lazy..... Or are you one of those people who are looking for a grail and want to go to an island for a holiday and a neural network will work?

I am one of those who likes to rack their brains and who is not looking for a grail. Forex is just a hobby for me. I'm not a trader, I'm a programmer and I'm just mastering a new area, where I have the opportunity to apply my knowledge.
 
klot:

NeuroShell2 is a great program, but in order to apply it to the market you need to add a binding to MT4. This, of course, is not difficult, but it is time-consuming... In Neuroschell Day Trader all studies and hypothesis tests can be done in 2-3 days. And you can check everything in real time in visual testing mode - a unique feature! You can create any NS and test it immediately. ...

But I've got some difficulties with it - I haven't found how to implement Kohonen's network, although NeuroShell2 has it. I wanted to try them out, as the research of classification results in Deductor inspires some optimism. In general, Deductor has friendlier interface, unlike NeuroShell or Daytrader, that hasn't been touched by interface designer's hand :-) Can you tell me, where are the Kohonen nets buried in Daytrader? Maybe for this need to put some kind of add-on, because search for the same helpe did not give anything - about them there is not a word.
 
TedBeer:
klot:

NeuroShell2 is a great program, but in order to apply it to the market you need to add a binding to MT4. This, of course, is not difficult, but it is time-consuming... In Neuroschell Day Trader all studies and hypothesis tests can be done in 2-3 days. And you can check everything in real time in visual testing mode - a unique feature! You can create any NS and immediately check it.

I'm having trouble with it - I couldn't find how to implement Kohonen's network, although NeuroShell2 has it. I wanted to try them out, as my research on the classification results in Deductor inspires some optimism. In general, Deductor has friendlier interface, unlike NeuroShell or Daytrader, that hasn't been touched by interface designer's hand :-) Can you tell me, where are the Kohonen nets buried in Daytrader? Maybe I should install some addon, because search for help did not give anything about them, not a word.


NSDT has no Kohonen, but there are other classifiers in the Adaptive Net Indicators addon. Although it is possible to connect NeuroShell2 net directly to MT4 or NSDT.

How do you plan to use Kohonen's network for trading? It's not three classes, but much more. You should think of some algorithm to parse the classes.

123456789101112
New comment
Reason: