Advisors on neural networks, sharing experiences. - page 2
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
That's why you were suggested to switch from buy to sell.
I have not heard of anyone making stable money on neural networks yet, it is very unstable. The simplest TS on muwings are more stable.
There is little information on the forum about ready-made solutions, and the effectiveness of neural networks for trading in the market. I suggest discussing and sharing experience here. If there is already a thread with discussion, please link to it.
I am using classes from here, simple multilayer Perspectron. I hope classes count correctly, I am counting on the author's experience. Started experimenting, interesting :)
At the moment I have 11 indices on input, output is a zigzag, shifted by 1 bar into the future.
These are the results of my 8 month grid. I am training on 1000 bars, 10000 epochs, 70 neurons in a hidden layer. Signals are reversed and purely by neural network, without additional filters. 15 min tf.
Try it on 8 years.
The network is well trained for a certain area, and if you do everything according to fancy, then you should have a sample to stop training, the so-called test sample, so training is also slightly adjusted to it. Is the graph shown on the validation section at least or is it a training section?
Try it on 8 years.
The network learns well for a particular plot, and if you are doing it right, you should have a sample to stop learning, a so-called test sample, so the learning adjusts slightly for that too. Is the graph shown on the validation section at least or is it a training section?
There is not much information on forum about ready-made solutions and effectiveness of neural network for market trading.
The neural net itself does not generate ideas. All it does well is adjust to the story on a particular section of the story.
Now a question: What is the difference between the signal from your 10 indicators and the signal from the neural network? And in continuation: What do you plan to get from the output of the neural network? No slogans like "perfect signal" .....
The neural net itself does not generate ideas. All it does well is adjust to the story on a particular section of the story.
Now a question: What is the difference between the signal from your 10 indicators and the signal from the neural network? And in continuation: What do you plan to get from the output of the neural network? No slogans like "perfect signal" .....
It's not about fitting to a story, it's about learning some patterns. If the pattern found is stable, it will work for some time in the future as well.
I don't understand what the difference is and what signals are meant. I plan to get an output, as strange as it may sound :) That is a specific solution, in my case it is a forecast for 1 bar ahead.
Yes, this is a validation, learning vector of only 1000 bars.
OK.
I have come to the conclusion in my time that training for a small segment is fraught with unpredictable network behaviour on future bars. The market may fall in the training, and then let's say it starts wobbling up and down. And then the net will start pouring. But if you build such a net that, say, in 10 years of training and in 5 other years shows a profit, that's an achievement.
OK.
I have come to the conclusion in my time that training for a small segment is fraught with unpredictable network behaviour on future bars. The market may fall in the training, and then let's say it starts wobbling up and down. And then the net will start pouring. But if to build such a net that will show profit on, say, 10 years of learning and 5 other years, then it is an achievement.
For that we should add OpenCl to this class, that such huge samples could be calculated, because one core and 1000 bars will take more than 5 minutes :) But so far I haven't figured out how to transfer the calculations to the video card.
What did you feed into it and what was the network structure?
It's not about adjusting to history, it's about learning some patterns. If the pattern found is stable, it will work in the future for some time.
I don't understand the question, what is the difference and what signals are meant. I plan to get an output, as strange as it may sound :) That is a specific solution, in my case it is a 1 bar forward forecast.
Are you aware that all patterns lag relative to linear indicators and in this case the neural network is useless.
Regarding the second question .... Who gave you the idea to do neural networks if you are "floating" in basic questions, if not drowning....
You are aware that all patterns lag relative to linear indicators, in which case a neural network is useless.
To the second question .... Who gave you the idea to do neural networking if you're not drowning in basic questions....
You're making me uncomprehending.) What do you mean that all patterns are lagged relative to linear indicators? I meant that the combination of indicators on the inputs forms a pattern which the grid learns by adjusting the weights to obtain the correct output. What kind of lag are we talking about here?
Nobody told me, I just found it interesting.)
You are aware that all patterns lag relative to linear indicators, in which case a neural network is useless.
On the second question.... Who gave you the idea to do neural networks if you are "swimming" in the basic questions, if not drowning....
I'll look for my topic about neural net.... It was on 4 and a long time ago.
http://forum.mql4.com/ru/38550
The input was the price difference with some lag (order of a few hours). The output is a forecast for several hours ahead binary plus or minus.
All nets were obtained by architecture brute force.
The main problem was gluing multiple forward tests. This should be automated in a good way.