Using artificial intelligence at MTS - page 24
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
A question has arisen. Does anyone have any criteria on how to determine whether a cochonen net is trained or not.
If 80-95% of trades are profitable according to NS signals, then we can say that the network is trained.
I have a question. Does anyone have any criteria to determine if the cochonen network is trained or not.
If 80-95% of trades are profitable according to NS signals, we can say that it is trained.
In the case of Kohonen maps, this is still a long way off. If we take a small map of say 50x50, we will get 2500 classes of possible outcomes. We still need to come up with an algorithm to make trading decisions....
I have a question. Does anyone have any criteria to determine if the cochonen network is trained or not.
If 80-95% of trades are profitable according to NS signals, then you can say that you are trained.
In the case of Kohonen maps, this is still a long way off. If we take a small map of say 50x50, we will get 2500 classes of possible outcomes. You still need to come up with an algorithm to make trading decisions....
I have it simple. I code candlesticks according to Likhovidov's method.
I train the network on random number sequences generated by the sensor.
The third step, finding entry and exit points.
The same was used in the contest but it was fitted by history.
Just when training the network, I output the deviation of the input array from the vector of weights. It turns out quite a big difference. (That's what I think) and it doesn't decrease between the maximum and minimum deviation per epoch. I tried different actions, but the result is the same. And hence the question of the learning criteria.
The result is 2 to 1, for two profitable trades one is losing.
Here is a question. Does anybody have a criterion - how to determine if the Kohonen net is trained or not?
If 80-95% of trades are profitable according to NS signals, then we can say that it is a trained network.
In the case of Kohonen cards this is still a long way off. If we take a small card e.g. 50x50, we will get 2500 classes of possible outcomes. We still need to come up with an algorithm to make trading decisions....
I've got it simple. Coding the candles according to almost Likhovidov.
The network is trained on the random number sequences generated by the sensor.
The third stage, finding entry and exit points.
The same thing went for the contest, but the tnm was fitted by history.
Just when training the network, I output the deviation of the input array from the vector of weights. It turns out quite a big difference. (That's what I think) and it doesn't decrease between the maximum and minimum deviation per epoch. I tried different actions, but the result is the same. And hence the question of the learning criteria.
Learning result is 2 to 1, for two profitable trades one is losing.
Here is a question. Does anybody have a criterion - how to determine if the Kohonen net is trained or not?
If 80-95% of trades are profitable according to NS signals, then we can say that it is a trained network.
In the case of Kohonen cards this is still a long way off. If we take a small card e.g. 50x50, we will get 2500 classes of possible outcomes. We still need to come up with an algorithm to make trading decisions....
I've got it simple. Coding the candles according to almost Likhovidov.
The network is trained on the random number sequences generated by the sensor.
The third stage, finding entry and exit points.
The same thing went for the contest, but the tnm was fitted by history.
Just when training the network, I output the deviation of the input array from the vector of weights. It turns out quite a big difference. (That's what I think) and it doesn't decrease between the maximum and minimum deviation per epoch. I tried different actions, but the result is the same. And hence the question of the learning criteria.
Learning result is 2 to 1, for two profitable trades one is losing.
Here is a question. Has anyone has a criterion - how to determine if the kohonen network is trained or not.
If 80-95% of trades are profitable according to NS signals, then you can say it is a trained one.
In the case of Kohonen cards this is still a long way off. If we take a small card e.g. 50x50, we will get 2500 classes of possible outcomes. We still need to come up with an algorithm to make trading decisions....
I've got it simple. Coding the candles according to almost Likhovidov.
The network is trained on the random number sequences generated by the sensor.
The third stage, finding entry and exit points.
The same thing went for the contest, but the tnm was fitted by history.
Simply, when training the network, I output the deviation of the input array from the vector of weights. It turns out quite a big difference. (That's what I think) and it doesn't decrease between the maximum and minimum deviation per epoch. I tried different actions, but the result is the same. And hence the question of the learning criteria.
The result of learning is 2 to 1, for two profitable trades one is losing.
I wrote above about the learning criterion... - it's true!
In general, try using Neuroshel2, it has a classic example of Kohonen cards. You should try it and a lot will become clear.
And on Tartan's forum I placed a complete neural network library for MKL4, with 5 neural network algorithms implemented, including Kohonen Maps.
I've got an example of neural network training algorithm (written purely in MQL4) and trading strategy parameters searching in one function - it's like auto-optimizer.
What is this Tartan forum?
What is this Tartan forum?
Google rus "forex tartan neuro"
http://www.fxexpert.ru/forum/index.php?showtopic=656?
Folks, please advise how.... during optimization for a certain time interval different results are obtained.... it means that the genetic algorithm each time selects a new path of genetic development :-). Anybody faced such a nonsense? Because of these problems it is impossible to get statistics on strategy success....