Even newbies can contribute:eek:
Start thread on Neural Nets 101. Any way, we can still catch some pips:D
You can use this develop your srategy then develop your NN to do just that
I have been reading this topic with great interest. Quick question(s) however:
Why are you using lagging indicators to forecast priceaction?
If you are using the EMA to forecast price, doesn’t it predict the current price since it is persistantly using data from 15 minutes ago?
Also, can this nn be converted to a H1 timeframe?
Final question, where do you keep the data thet the system has thought? And is it retrainable, with different, more current or historical data. As you know last year was dollar beating year and this year was recession, so very limited data as patterns have been pretty much the same = big move up.
Thanks in advance.
Many traders apply neural nets incorrectly because they place too much trust in the software they use all without having been provided with proper instructions on how to use it properly. To use a neural network the right way and, thus, gainfully, a trader ought to pay attention to all the stages of the network preparation cycle. It is the trader and not his or her net that is responsible for inventing an idea, formalizing this idea, testing and improving it, and, finally, choosing the right moment to dispose of it when it’s no longer useful. Let us consider the stages of this crucial process in more detail:
1. Finding and Formalizing a Trading Idea
A trader should fully understand that his or her neural network is not intended for inventing winning trading ideas and concepts. It is intended for providing the most trustworthy and precise information possible on how effective your trading idea or concept is. Therefore, you should come up with an original trading idea and clearly define the purpose of this idea and what you expect to achieve by employing it. This is the most important stage in the network preparation cycle. (For related reading, see Lessons From A Trader's Diary.)
2. Improving the Parameters of Your Model
Next, you should try to improve the overall model quality by modifying the data set used and adjusting the different the parameters.
Figure 1: Specifying the optimization algorithm and its properties
3. Disposing of the Model When it Becomes Obsolete
Every neural-network based model has a life span and cannot be used indefinitely. The longevity of a model's life span depends on the market situation and on how long the market interdependencies reflected in it remain topical. However, sooner or later any model becomes obsolete. When this happens, you can either retrain the model using completely new data (i.e. replace all the data that has been used), add some new data to the existing data set and train the model again, or simply retire the model altogether.
Many traders make the mistake of following the simplest path - they rely heavily on and use the approach for which their software provides the most user-friendly and automated functionality. This simplest approach is forecasting a price a few bars ahead and basing your trading system on this forecast. Other traders forecast price change or percentage of the price change. This approach seldom yields better results than forecasting the price directly. Both the simplistic approaches fail to uncover and gainfully exploit most of the important longer-term interdependencies and, as a result, the model quickly becomes obsolete as the global driving forces change.
Conclusion
You will experience real success with neural nets only when you stop looking for the best net. After all, the key to your success with neural networks lies not in the network itself, but in your trading strategy. Therefore, to find a profitable strategy that works for you, you must develop a strong idea about how to create a committee of neural networks and use them in combination with classical filters and money management rules.
note: not my words but i agree with them. i cant post links.
i had some success with my system. i will start posting more result after i finalize everything.
Hi Aroon,
Thanks for your great job, I have a comment regarding the output values of your writer:
you are using
output[0] = EMA5(1)-EMA5(2)
output[1] = EMA5(2)-EMA5(3)
output[2] = EMA5(3)-EMA5(4)
This will predict 3 different EMAs(at different shifts), I suggest your output to be:
output[0] = EMA5(1)-EMA5(2)
output[1] = EMA5(1)-EMA5(3)
output[2] = EMA5(1)-EMA5(4)
This will predict 3 values for the EMA at the same shift, allowing for more precise prediction if you average the values.
One more thing about FANN, I have a CLI fann tool written by me, running on Windows, if you are serious about using FANN tool, I can post the EXE here.
Regards,
"This is an efficient system of scalping that works in 1 minute up to 1 day all periods and all Currency
Ema 3
Ema 5
Ema 7
Ema 9
Ema 11
Ema 13
Color yellow
Ema 21
Ema 24
Ema 27
Ema 30
Ema 33
Ema 36
color Green
Ema 55 Color Red
Well
The system in The Oanda Forum.
And Way Works is
Buy Or Sell When the Group of Ema`s Yellow Breaks Ema 55
Take Benefit When Yellow Group Touches The Group Green.
If The Yellow group does Pull Back in Ema 55 or In Green Group Buy or sell again
Take Profit 10 Pip in Breaks Ema 55 And take Profit 5 pips In Pull Back
Stop loss 5 Pip 5 Minute Periods
The Good thing is Works with All periods and all the Currency
Arun.
I reinstalled all the indciators and now it’s ok. It’s seems very promising.
Now I am trying to install the signal script you made but I cant see it runs anywhere.
What’s wrong?
Hi,
You may be right regarding the outputs. NeuroTrend was an initial trial to show that it is possible to create a leading indicator using Neural Networks.
I no more predict 3 future values but only one future EMA.
Thanks for the offer. Currently I am fine with JNNS and if in future decide to use FANN I can try your tool.
Thanks,
Arun
Its an expert and not a script file. Please run it in strategy tester. It should work fine.
@ALLLLLL,
[B]Since I am involved in a couple of projects, I am not able to find time to actively contribute to this thread. I have decided to implement the design we have discussed once I am done with my projects (which includes also optimizing NeuroTrend and building a reliable system on it).
I would really appreciate if someone wants to go ahead and implement the discussed NN design. I have given the overall concept and all necessary files to device, train and use a neural network based indicator. I can give support in case of difficulties with NeuroTrend and also suggestions for new implementations based on my experience with NeroTrend and NN in general.[/B]
Best Wishes,
Arun
Arun and All,
I have also been very busy with other projects and I won’t be available for some time neither. The proposed system should be easy to try for someone who has coding experience.
I will get into it later (hopefully this year!!) for sure.
Hi Arun,
I wish to see you soon here again.
I ran the file on the strategy tester as you wrote but I could not see any results. What did I do wrong?
Hi sam123,
Could you briefly explain me what did you try with the expert file. I can better help if you can explain in short what you want to do with it.
Use the expert to generate (in other words export) the signals to a file. After you run in the strategy tester it will write the signals to a file which you can find in /tester/files folder. You can then load this file in JNNS to train a network.
Cheers!
Arun
does this work only on 15 min charts?
thnx
Arun.
-
I thought it is a regular ea. After I read your explanation in the last post I realized that I was wrong.
-
I found the txt file in the files folder.
In fact, there are 4 txt files there. -
I wish to go on forward with these files but I don’t understand what to do.
What is “loading in JNNS to train a network”.
Please explain it as you explained in the last post so I will understand easily and will be able to help this system to work fine.
Thanks.
Hi AruunThank you for all your contributions here…I’m now starting to understand this system…i have tried to created trainning set using trainningset creator you post…and sucsessfully export to JNNS…But I don’t know how long or how many cycle required for trainning? What is maximum error sould be reach to get accurate prediction…Because, with the net file result, I didn’t get same indicator result as yours…
thank you in advance for your help
Regards
rohyani
I have the same problem… What should be the parameters/functions/weights to train the network based on the pattern file.
Thanks
Hi,
Sorry for this late reply. There is no clear answer for the question cycles and error limit. In ideal case the error should reduce to zero. In practice this is never the case. My experience says find the global minima for the validation curve. This will give you the best results. There is also no fixed rule for the cycles. I usually start with 10,000 cycles and first identify the cycles for the global minima. Then I retrain for the noted number of cycles and save it. In NeuroTrend I used tanh as activation function. Learning function can be selected as per your experience. In my experience Rprop is the best but not for all kinds of networks. For NeuroTrend yes 
All the best,
Arun
Why didn’t see the attatched?thanks!
The file is attached in the first post
http://forums.babypips.com/attachment.php?attachmentid=4704&d=1218051436
I see it,thank you.Let me tryit.
sorry,my English is poor.:mad: