Advisors on neural networks, sharing experiences. - page 5

 
Stanislav Korotky:
Have done some things in the past. You can read it on the blog under the forecasting tag.
Interesting, you just used the notion of chaotic processes I commented on above. But it's so fucking complicated for me :)
 
Maxim Dmitrievsky:
Yeah, I've read it... lots of philosophy... few realizations... so people like to philosophize more than to do something real :) :) :) Just kidding. But I would like to see something that would stably trade at + on neural networks. Without code and algorithms disclosure, just as a motivating factor :)
.............
If anything, I would be glad to have a look... No kidding. But please do not put Mr Reshetov here... :-)
Vaughn. The whole Mr. Batter seems to have quit neuronics because of the sinking...
 
There is a website like neuroproject.com. I haven't been there for a long time. And there's an article here, too: Type in Fann Fann Neural Network in English and read it, where neuro is connected to indicators. I got into the subject a few years ago... then I switched to other things...
 
Serqey Nikitin:
1

It is the same as identifying patterns in the market, in which case a grid is definitely not needed.



2

Still, training the grid in a particular area is its disadvantage, and it will have to be retrained periodically, despite all the cooking skills ...

1 I agree.
2 Then how do you train her?
If not in a certain area ...
To retrain it according to the recommendations for optimising any other non-neuro ts.
 
Maxim Dmitrievsky:
Standard question, what were you trying to teach? ))

My approach to this question is only the prices themselves (their differences) and the variables preferably independent at most.

As for rationing (bringing to the interval), of course, each variable by its minimax, not by any one variable.

 
Алексей:

My approach to this question is only the prices themselves (their differences) and the variables preferably independent at most.

As for rationing (interval reduction), of course, each variable by its minimax, not by any one variable.

And how do you normalise prices? That is, I'm wondering when prices go outside the training range, say in a trending market, how does that affect the network results? Ah, the price difference... OK, got it.
 
Алексей:

My approach to this question is only the prices themselves (their differences) and the variables preferably independent as much as possible.

What are the variables? And how are you getting on by the way? )
 
Комбинатор:
What are those variables? And how are you doing by the way? )
I'm making good progress, but modestly. I have not finished the real test yet. I'll be sure to post it later.

I make a model in parallel. The idea is old and simple, it has long been picked up. I select a combination of lags, for which I take the price differences so that the resulting skewness in predicting the sign of growth is a) statistically significant and reproducible on an independent sample, and b) so that there is as little redundancy between these lag-variables as possible. There are already so many intricacies, all on the theory of inf. Observing the requirement for independence of observations. Maybe later I will draw it up in some generalized way. But I have not discovered any miracles yet, frankly. ) I am running 5M minutes and generating rules. Weak dependencies are quite stable, but with small MO yet.

But I have everything without neural networks. I finally am able to obtain easy to read rules and use them to create Expert Advisor code. All thanks to the capabilities of theoretical information.

 
Roman Shiredchenko:
1 I agree.
2 Then how do you train her?
If not in a certain area...
Re-train it according to recommendations for optimizing any other non-neurotz.
It should determine the period of retraining by itself, and it should have several trained periods and several trading strategies. This will allow you to catch several patterns at once, and when one starts to fail, the other just starts to earn. And it should be retrained in real time.
 
Maxim Dmitrievsky:
Standard question: what were you trying to learn? ))

self-learning neural network.

You put indices, a graph, mark entry and exit points as you see them.

The grid seems to be looking for patterns or remembers the position and direction of indices and accounts +-% deviation from the ideal.

i did not learn a thing on gold.

Maybe you should have taught it on Eurobucks...