Neural network - page 11
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
It can only be assumed that the numbers of the squares in which the price is placed are fed in
Rather, the numbers of the top and bottom square from each column that are covered by the price. In that case, the discreteness of the vertical breakdown is unclear.
UPD: I mean, not the discreteness itself, but the step of the split.
I lied a bit about the 2 output options. There are more of them. But a lot of them are the same to four decimal places.
I don't understand about the output classifier. Where can I read about it?
What type of networks do you use? With what type of nets did you manage to achieve the minimum error on the test sample? Which of them showed the best results in trading?
to marketeer && StarBars: I train the grid with indicator similar to MPP (from Ivanov's diploma) smoothed with 6-period moving average. I read that evenly distributed teaching values are good for the network. I was a little wrong about the thresholds. Not a threshold, but a sieve of unnecessary signals using the X variable. If output>0.5+X, we buy; if output<0.5-X, we sell. Thus, "weak" signals lying between 0.5-X and 0.5+X are eliminated. Do you not use such a lobuda?
I lied a bit about the 2 output options. There are more of them. But a lot of them are the same to four decimal places.
I don't understand about the output classifier. Where can I read about it?
What type of networks do you use? With what type of nets did you manage to achieve the minimum error on the test sample? Which of them showed the best results in trading?
Thresholds often increase the percentage of correctly detected patterns (pwin), but these patterns are exactly where the price (benefit) and some smoothing filter diverge, in short thresholds do not help anything, although the pwin increases...
I'm a bit confused, it's more about target functions, but in any case some construction is applied, rules which produce the final signal...
Tsaregorodtsev's website, his articles.
Almost everything I need I get from MLP, in rare case NS2->Polynomial Net.
Trade result and architecture are not as strongly linked as input - error - trade, for example. As a rule, you can win no more than 1% on architecture, and in rare cases, in which I meet them more often, I have to optimize the structure of my grid...
to StatBars:
Is pwin the winning percentage?
Please tell me, so that the arrows in the picture below are not located at the same interval, just use the threshold (or "eliminator") about which you wrote above? That is, where there are no arrows - is it Out of network hit in 0.5-X or 0.5+X?
to StatBars:
Is pwin the winning percentage?
Please tell me, so that the arrows in the picture below are not located at the same interval, just use the threshold (or "eliminator") about which you wrote above? That is, where there are no arrows - is it Out network hit in 0.5-X or 0.5+X?
I don't understand the question... Is it still relevant?
If so, please elaborate...
In general, if the signal is saved, the arrow is not copied, otherwise every bar will be some arrow, which is not convenient... If the signal has changed, then on the same bar you put the appropriate arrow...
about pwin - yes.
Dear guru, please help me to understand neural networks as long as I try, I still can not understand how they work, how to make them, how to train them, if it is not difficult, please show simple examples explaining what and how....
I am interested in the following questions:
1. As far as I understood each neuron of the network is the same function... but I don't understand how one and the same function with the same data can give different values...
2. How to normalise quotes without knowing their minimum and maximum???
I would like more graphical information and at least simple examples of neural network with a built-in learning mechanism...
Thanks in advance hope for your help...
Articles won't help. You have to understand the maths.to StatBars: About the arrows I understand, thank you. I don't get it, are you using a eliminator? How many output variants should a 3 layer neural network produce? For some reason my network reaches minimum error on test sample when there are only two variants. Because of this, I can't get this sifter to work. Outputs constantly point to trade actions and never to waiting. I use all sorts of teacher values (much more than two) when training. Do you know what might be the reason? How many values is better to use when training a backspreading network? Do the outputs of each neuron in the network need to be passed through the activation function?
Is the sill screen a sill screen? - No I don't, at the same time that doesn't mean you shouldn't use them.
//----------------------------------------------------------------------------------------------------------------------------------//
You can have as many output signals as you like, and you should have as many as you need, or you can design a separate grid for each signal you need.
//----------------------------------------------------------------------------------------------------------------------------------//
I think your Action - No Action - No Action classes are unevenly distributed, there are very few examples of the No Action class compared to the others. Perhaps the reason is something else, throw a training sample here.
//----------------------------------------------------------------------------------------------------------------------------------//
The values of what outputs, inputs?
Neuron itself sums and passes through a sigmoid, and it is better to leave only connection weight between neurons...
Yeah, we can discuss it endlessly but never come to a sensible result......