Better NN EA development - page 25
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
There are two good ways to develop NN：
1. Use of NN software such as NeuroSolutions,NeuroShell or Matlab,I found Barnix in July last year have studied the NeruoShell.I studied the NeuroSolutions, and found that the software is not easy to learn.This software has a lot of built-in NN model, such as PNN.Although I had done a complete test use RBF/PNN, but it has too many details need to grasp.This is a little difficult to a novice like me.
2. Who can provide a complete PNN source code? for the people who understand programming may be a faster and better way of development strategy based on PNN.
Barnix and Stef can talk about the content of this area?Certainly a lot of listeners interested, please?
PNN source code
Free forecast program (statistically learning self-organizing neural networks)
Home | Sciengy.com
PNN source code
Barnix your information is really rich, you're a enthusiastic person, very grateful!
Simply read the source code, I am prepared to try to convert it to c#,If someone has intended to do it so, we can do division.
The Art of learning
Hello Supervisor of NN,
i'm developing alot with ANN, SOM GA, GP and Reinforcement Learning. do this in gameprogramming, datamining and just for fun.
Now i like to use forex as a testbed for artificial intelligence.
I have to learn what the AI have to learn. my thoughts:
if i started with the basics i tried to learn a simple perceptron with an amount of inputs to give me an output sell or buy. it failed.
well, i thought: "What informations are important for the AI to learn?"
i don't want a direct prediction NN. a simple Probability NN would be enough. But what should the inputs be? First i started to collect Lows, Highs, Opens and Closes of every bars in history and fed the PNN to give me a percentage where the next candle will close. i failed. why? it is not possible to predict the market. too many happenings influences the market. there are news, fear and greed etc. the AI could not read the minds of investors and the big dogs.
the future of the forex market is written in the moment of now. (only you have secret informations, then you do not need AI)
i tried several indicators for inputs, uses klot ZIG-ZAG learning with MFI. but failed also. Why? this indicator uses the volume. and volume is the worst indi ever. this are my thoughts.
i begin with the next step. classification. i tried SOM for pattern classification. but failed. because if the SOM recognize a developing pattern it will change in the next minute, because it forms another pattern. and if it found a ready made pattern then it is over and a next pattern begin to develop.
so, how to find the way of market prediction?
i think it will work with a kind of a higher AI. what does this mean? we need a collectiv system, which work not only on a single pc system. we need the power of all NN's, PNN's, SOM's and very important is reinforcement learning on every single pc. connected in a hughe network of participants.
iit is like filesharing. but we share inputs for a collective cluster system. every trader has its own part in this system.
but first i will try to code a reinforcement learning EA.
this would be my stage number 3.
some thoughts from you?
regards PriNova.
P.S.: to BigBoppa. i tried to register at the german forum eaforex. but get no answer until now. is this forum dead? maybe you know the admin and could send him a message
so, how to find the way of market prediction?
Stop trying to teach an NN to predict. Instead, teach it how to trade. There is a difference.
Stop trying to teach an NN to predict. Instead, teach it how to trade. There is a difference.
does this mean that every nn system needs trading rules?
the NN is like a filter for a specified trading rule. for example rule:
buy if stoch signal breaks level 20 upwards and vice versa. and the NN filters the right signals because of pattern recognition of the history and weight adjustment is made after lossing trade to filter them out. right?
P.S.: to BigBoppa. i tried to register at the german forum eaforex. but get no answer until now. is this forum dead? maybe you know the admin and could send him a message
I will tell the admin...
, btw, do you speak german?
bbop
yes, i speak german and searching for guys, which like to participate in my project: reinforcment learning combined with AI and MT4. client cluster networks etc.
for interests please PM
PriNova
Stop trying to teach an NN to predict. Instead, teach it how to trade. There is a difference.
With this, I so very strongly agree...
In my opinion, any predictions can only provide us with "possible" targets - it is up to the market to agree with these target or not.
As such, you still need a strong trading strategy in support of your prediction engine.
So many developers (including myself) spend years trying to find the ultimate prediction engine. This is fun, challenging and exciting all at the same time. But remember - in the end (just like in the casino) you will pay for having a good time.
In this search for the ultimate prediction engine we loose focus of what is really important - balance in the bank. The best advice I ever saw on some forum was...
Trading should be boring.
Because - if you find it exciting etc. you loose focus of what is important -your account balance.
Thanks to Barnix for all his hard work - he truly is pushing the envelope (I think) and I really enjoy and appreciate his work.