THE IDEA EXCHANGE - page 10
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
Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. In principle, we can optimize it by limiting parameters with the same set of indicators.
Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, you can optimize by limiting parameters using the same set of indicators.
The contest uses a Kohonen layer of 250 neurons. We have to do about 1500 of them. It will take about 100 machine hours to train neuronics, maybe more. Maybe even faster, if the training algorithm is different (faster). For 250 neurons it takes around 10 hours. But the problem starts with training not the mesh, but your Expert Advisor. Here is the problem. It took me three weeks to train my Expert Advisor, but to put it crudely. In fact, something was changing all the time. Net time - eight to ten hours. If I have 1500 neurons, it will take me about 80-120 machine hours to train the Expert Advisor. But it's only for one currency. And we need to do as many of them as possible. My resources are not enough for that. Even if I change computers.
I recommend reading Neurocomputer Science: Theory and Practice by F. Wasserman. It is very well written. If you need it, I will send it to you by e-mail. I can send you not only this one, but other books as well.
Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, we can optimize the grid by limiting parameters using the same set of indicators.
Somewhere here (in the forum) I posted an advisor Not kohonen but multilayer nets
It seems to work as a fish
Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, we can optimize the grid by limiting parameters using the same set of indicators.
Somewhere here (in the forum) I posted an advisor Not kohonen but the nets are layered
It will do as a fish.
Can you be more specific. Apparently it has passed me by.
Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, you can optimize by limiting parameters using the same set of indicators.
The Kohonen layer of 250 neurons is used in the contest. We should make about 1500. It will take about 100 machine hours to train neuronics, maybe more. Maybe even faster, if the training algorithm is different (faster). For 250 neurons it takes around 10 hours. But the problem starts with training not the mesh, but your Expert Advisor. Here is the problem. It took me three weeks to train my Expert Advisor, but to put it crudely. In fact, something was changing all the time. Net time - eight to ten hours. If I have 1500 neurons, it will take me about 80-120 machine hours to train the Expert Advisor. But it's only for one currency. And we need to do as many of them as possible. My resources are not enough for that. Even if I change computers.
I recommend reading Neurocomputer Science: Theory and Practice by F. Wasserman. It is very well written. If you need it, I will send it to you by e-mail. I can send you not only this one but other books as well.
Thank you, Victor. I think it will be useful to review it. My email address is fxfion(dog)mail(dot). ru.
My code has some implications, i.e. i get some values for the indicators, but i don't understand the structure in general, i messed up with arrays.
Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, we can optimize by limiting parameters using the same set of indicators.
The Kohonen layer of 250 neurons is used in the contest. We have to make it about 1500. It will take at least 100 machine hours to train the neuronics, maybe more. Maybe even faster, if the training algorithm is different (faster). For 250 neurons it takes around 10 hours. But the problem starts with training not the mesh, but your Expert Advisor. Here is the problem. It took me three weeks to train my Expert Advisor, but to put it crudely. In fact, something was changing all the time. Net time - eight to ten hours. If I have 1500 neurons, it will take me about 80-120 machine hours to train the Expert Advisor. But it's only for one currency. And we need to do as many of them as possible. My resources are not enough for that. Even if I change computers.
I recommend reading Neurocomputer Science: Theory and Practice by F. Wasserman. It is very well written. If you need it, I will send it to you by e-mail. I can send you not only this one but other books as well.
Thank you, Victor. I think it will be useful to review it. My email is fxfion(dog)mail(dot). ru.
My code has some implications, i.e. i get some information about the normalization of indicator data, but i don't understand the structure in general, i messed up with the arrays.
I sent it to you. There are many more interesting things. I may have a look at klota's developments on the spider. Although I don't like all of them. And regarding arrays - all neuronics is arrays and nothing else. It simply matters what we do with these arrays.
About the Kohonen maps. I don't use them, they are needed, I think, only for visualization - and that should be done in a suitable software. And the Kohonen layer solves the same problem.
I got acquainted with the NS on the whole and realized that we do not always trade with our brains successfully, and it would be hard to train an "artificial brain". So far, at least I have not heard any enthusiasm about the use of NS, maybe the market is too tough for them?
If you do not know the difference between the market and the real one, then you do not know how to use the artificial brain. So far, at least I have not heard any enthusiasm about the use of NS, maybe the market is too tough for them?
Not really. It is not about brains. It's about the problem statement. Not every problem can be solved by a neural network. But a lot of problems can be solved. Like pattern recognition, information compression. There are a number of other tasks that it handles with success. In the beginning I tried to predict the High and Lov of the next day. For values below the average the accuracy was about 80%, if higher - 5%. Forex does not fit the normal distribution law. I have to convert values beforehand. But the result cannot exceed the maximum value of the grid. Although many people have gone this way, as I noticed. But Mr. Reshetov "helped" a lot with the neuron. I fell for it myself and spent three months before the contest for his solutions. Although it helped me in some way. I created a mechanism for training my Expert Advisor with neuronics. I do not want to say that it is perfect. But it helped me a lot. And thanks to kandid, for his article in response to my question.
I recommend reading F. Wasserman Neurocomputing: Theory and Practice. It is very well written. I can send it to you by email if you need it. I can do other books, and not only this one.
If it is not difficult, I need to do it too. My address is in my profile.
I have recently come to the conclusion that without NS my system cannot be taught to trade correctly. As I have seen I am a bad teacher. :-) I have an idea, that it needs proper clustering of data, with which my system works. Well, as far as I understand, they can be clustered using a Kohonen network. But my first attempts to get through all this have not yet led to any results. I know too little about it. I need to read something good that combines both clearly stated ideas and good examples of practical use.
I've read the whole cloth thread on NS, but it's not my level. I need to fill the gaps urgently.