Build Neural Network Indicator in MT4 using Neuroshell
Hi Akuma,
It is part of Neuroshell Run-time server, you can get it from ward, it is part of AI-Trilogy. You can use it to utilize trained net in NS Predictor/Classifier to other program such as Pascal, C, Delphi or Metatrader. If you already have NSpredictor/classifier then you need to buy only run-time server, sure they will check whether your already bought NSpredictor/classifier from them or not.
See http://www.wardsystems.com/rtserver.asp
regards
It is part of Neuroshell Run-time server, you can get it from ward, it is part of AI-Trilogy. You can use it to utilize trained net in NS Predictor/Classifier to other program such as Pascal, C, Delphi or Metatrader. If you already have NSpredictor/classifier then you need to buy only run-time server, sure they will check whether your already bought NSpredictor/classifier from them or not.
See http://www.wardsystems.com/rtserver.asp
regards
Hi Akuma,
Here is my true story..
When I come to this forum, I learned already about NSDT since I bought it 06/14/2006 but quite confusing and always failed to build trading strategy/prediction especially when apply it in forex market. You can see my post about noxa ... one of my own indicator..but still not getting success.
I thought I have made some mistakes, how good the tools that you use, it always depend on your way to use it and especially your personality/mentality. I am hunger to have a profitable trade without sufficient knowledge how to use my tools and mostly in practical the emotion cover all what ever strategy.
I changed the way I learn, as many user known, NSDT is a black box, you do not know how the prediction is made ...bla..bla..bla, then I began learning the detail about neural networks (buy some books or ebook), I found also that classic NS2 gave me better understanding about neural network rather than NSDT, how to select the output, input, training data range, NN configuration, etc. I tried also other software as comparison such as matlab, braimaker, neurosolution ..but I feel comfortable with NS2 even this is a very old software.
Using NS2 then I build some MT4 indicators and finally create my own EA (also bought some EA for learning), after feel confident that the EA is profitable then implemented it on real trading ..I started using EA just last February and asking my friend how to post it in myfxbook .. till today I still developing and will never stop..
So ..it was not an immediate one like fast food..need a lot of time, energy and money..
Here is my true story..
When I come to this forum, I learned already about NSDT since I bought it 06/14/2006 but quite confusing and always failed to build trading strategy/prediction especially when apply it in forex market. You can see my post about noxa ... one of my own indicator..but still not getting success.
I thought I have made some mistakes, how good the tools that you use, it always depend on your way to use it and especially your personality/mentality. I am hunger to have a profitable trade without sufficient knowledge how to use my tools and mostly in practical the emotion cover all what ever strategy.
I changed the way I learn, as many user known, NSDT is a black box, you do not know how the prediction is made ...bla..bla..bla, then I began learning the detail about neural networks (buy some books or ebook), I found also that classic NS2 gave me better understanding about neural network rather than NSDT, how to select the output, input, training data range, NN configuration, etc. I tried also other software as comparison such as matlab, braimaker, neurosolution ..but I feel comfortable with NS2 even this is a very old software.
Using NS2 then I build some MT4 indicators and finally create my own EA (also bought some EA for learning), after feel confident that the EA is profitable then implemented it on real trading ..I started using EA just last February and asking my friend how to post it in myfxbook .. till today I still developing and will never stop..
So ..it was not an immediate one like fast food..need a lot of time, energy and money..
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
Arryex
Is neurshell data sensitive? I made successfull strategy using feed from demo and when i try to implement same strategy using live data feed the results have changed .
I i use demo feed even after commision taken out i make decent money but i assume that neuroshell is data sensitive . If a staregy has been trained on eurusd 15 mins from one provider and if i use it on different data provider my results change.:innocent:
Is neurshell data sensitive? I made successfull strategy using feed from demo and when i try to implement same strategy using live data feed the results have changed .
I i use demo feed even after commision taken out i make decent money but i assume that neuroshell is data sensitive . If a staregy has been trained on eurusd 15 mins from one provider and if i use it on different data provider my results change.:innocent:
Hi Supremegizmo,
This is my personal opinion, any other advice are appreciated.
Previously I had the same problem, but the key solution for that is the selection of your input and output range. The nature of neural network is depend on the training data (most important, including other factors), as long as new data still within the range while you train there will be no change even for the new out of sample data or different trading pair.
Data pre-processing is the only way, using data distribution histogram I simplified the range of data. If you follow what explained in Brainmaker guide, you can also use cyclic analysis, standard deviation and sensitivity analysis.
By this technique, the training data can be simplified, removing unnecessary (garbage data), training the net in shorter time, and more robust during out of sample. Most of time I use excel to verify exported data from MT4 for analysis.
Within NS2, there are also some technique used in data pre-processing, i.e normalization [0,1], [-1,-1], <<-1,1>>, standard deviation for maximum and minimum calculation, etc.
As an analogy, If you train the NN only to learn a,b,c,d data, you can not expect NN to give a correct output when input become e. This common mistake that i did previously. In real example, you train NN only based on certain indicators for forex trade, this NN will not work whenever fundamental news (NFP, trade balance, etc) announced, why because we do not include fundamental factor on the trained net..
I made the same mistake and justify that my NN is not work or justify the software is bad, but actually I train the net with garbage data that is why the result also become garbage. Hence we should learn how to find which inputs to be used on the net.
Cheers
This is my personal opinion, any other advice are appreciated.
Previously I had the same problem, but the key solution for that is the selection of your input and output range. The nature of neural network is depend on the training data (most important, including other factors), as long as new data still within the range while you train there will be no change even for the new out of sample data or different trading pair.
Data pre-processing is the only way, using data distribution histogram I simplified the range of data. If you follow what explained in Brainmaker guide, you can also use cyclic analysis, standard deviation and sensitivity analysis.
By this technique, the training data can be simplified, removing unnecessary (garbage data), training the net in shorter time, and more robust during out of sample. Most of time I use excel to verify exported data from MT4 for analysis.
Within NS2, there are also some technique used in data pre-processing, i.e normalization [0,1], [-1,-1], <<-1,1>>, standard deviation for maximum and minimum calculation, etc.
As an analogy, If you train the NN only to learn a,b,c,d data, you can not expect NN to give a correct output when input become e. This common mistake that i did previously. In real example, you train NN only based on certain indicators for forex trade, this NN will not work whenever fundamental news (NFP, trade balance, etc) announced, why because we do not include fundamental factor on the trained net..
I made the same mistake and justify that my NN is not work or justify the software is bad, but actually I train the net with garbage data that is why the result also become garbage. Hence we should learn how to find which inputs to be used on the net.
Cheers
This is sample analysis I made in excel for exported data input NN from MT4.
Original data is about 58 K data, checking its data distribution, only data within -0.02825 to 0.034036 consist about 90% of data. then I remove the data out of this range. With this method, we can remove about 10% of data which having less frequency.
Original data is about 58 K data, checking its data distribution, only data within -0.02825 to 0.034036 consist about 90% of data. then I remove the data out of this range. With this method, we can remove about 10% of data which having less frequency.
Attachments
Another consideration I should mentioned here is I never use demo data to train my net why because most of broker has different data feed values between demo and real account.. I am not sure if this one of broker strategy ? ( sorry if you are a broker..he..he)
So I use data from real data feed..
My conclusion: NN is very sensitive to the data, to make your net more robust you should find the indicators which having values within the range (RSI, stochastic), i do not feed with OHLCV data because the range for these data is not fixed.
So I use data from real data feed..
My conclusion: NN is very sensitive to the data, to make your net more robust you should find the indicators which having values within the range (RSI, stochastic), i do not feed with OHLCV data because the range for these data is not fixed.
Hi Akuma,
Here is my true story..
When I come to this forum, I learned already about NSDT since I bought it 06/14/2006 but quite confusing and always failed to build trading strategy/prediction especially when apply it in forex market. You can see my post about noxa ... one of my own indicator..but still not getting success.
I thought I have made some mistakes, how good the tools that you use, it always depend on your way to use it and especially your personality/mentality. I am hunger to have a profitable trade without sufficient knowledge how to use my tools and mostly in practical the emotion cover all what ever strategy.
I changed the way I learn, as many user known, NSDT is a black box, you do not know how the prediction is made ...bla..bla..bla, then I began learning the detail about neural networks (buy some books or ebook), I found also that classic NS2 gave me better understanding about neural network rather than NSDT, how to select the output, input, training data range, NN configuration, etc. I tried also other software as comparison such as matlab, braimaker, neurosolution ..but I feel comfortable with NS2 even this is a very old software.
Using NS2 then I build some MT4 indicators and finally create my own EA (also bought some EA for learning), after feel confident that the EA is profitable then implemented it on real trading ..I started using EA just last February and asking my friend how to post it in myfxbook .. till today I still developing and will never stop..
So ..it was not an immediate one like fast food..need a lot of time, energy and money..
It's long run, Arry and I feel you're on right track now!
Look at your fxbook and I'm glad it's growing fast. I saw those's almost long trade and suprise it's survice Mar 15, the day's chart go down :-0
I'm beginning my journey :innocent:
Hi Akuma,
You should verify what you are seeing, if you are in the growth tab, the green bar on the chart is what profit in pips I get on the day, the equity and growth is in red line.
I should verify my net if my growth/equity is going down..on 14 Mar I get 519.15 pips while in 15 Mar I get 55.8 pips, it will be depend on market condition...
You should verify what you are seeing, if you are in the growth tab, the green bar on the chart is what profit in pips I get on the day, the equity and growth is in red line.
I should verify my net if my growth/equity is going down..on 14 Mar I get 519.15 pips while in 15 Mar I get 55.8 pips, it will be depend on market condition...
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
Arryex thanks for posting that information I will try it.
Apart from that my needs are something else , may be i am expecting too much from NST
let take example
I train a net to detect a head and shoulders pattern for usdcad 15 min chart . I understand now its has learnt and has saved this pattern in its memory.
Now i save this as a template say 'headandshouldersnet'
Now I wish to use this template on the oos sample of another currency then how do i make sure that the net uses its learned data to predict the h&s on other currency?
:clap:
Apart from that my needs are something else , may be i am expecting too much from NST
let take example
I train a net to detect a head and shoulders pattern for usdcad 15 min chart . I understand now its has learnt and has saved this pattern in its memory.
Now i save this as a template say 'headandshouldersnet'
Now I wish to use this template on the oos sample of another currency then how do i make sure that the net uses its learned data to predict the h&s on other currency?
:clap:
Hi Supremegizmo,
As you know our common issue is to find what is the output we need to find then the input need to be used. Maybe this technique may works or not, but you should try (no pain no gain right?)
- We identify the what is the core idea of h&s or other chart patterns, maybe the maximum and minimum values within a certain periods,
- How we can represent these values as a good output to be predicted? in order valid for all time period or even different pair. My previous technique used always to be a price difference or momentum or percentage and NOT the price it self, then use normalization.
- If you read the help explained in NSDT, they stated to use change or percent change, I follow this simple rule for stating the input as well:idea:.
Next, please present you progress so we can share and improve..Some people just issuing their question but hiding their progress result...
I will make my own research if you want about h&s pattern..of-course if you post something..:smart:
BUT do not assume that I already found a best strategy ..I am still finding and testing on it..
As you know our common issue is to find what is the output we need to find then the input need to be used. Maybe this technique may works or not, but you should try (no pain no gain right?
)- We identify the what is the core idea of h&s or other chart patterns, maybe the maximum and minimum values within a certain periods,
- How we can represent these values as a good output to be predicted? in order valid for all time period or even different pair. My previous technique used always to be a price difference or momentum or percentage and NOT the price it self, then use normalization.
- If you read the help explained in NSDT, they stated to use change or percent change, I follow this simple rule for stating the input as well:idea:.
Next, please present you progress so we can share and improve..Some people just issuing their question but hiding their progress result...
I will make my own research if you want about h&s pattern..of-course if you post something..:smart:
BUT do not assume that I already found a best strategy ..I am still finding and testing on it..
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
Arryex
Thx again for reply. I have pasting one of my strategies I made from NST. I already use the concept of normalisation for a robust system. The system I have attached is not the greatest but gives me some confidence that I have the ability to create winning systems.
I am a trader myself so i know what a net to output and if it doesnt outputwhat i want then something is wrong. For now I want my net to predict complex patterns which are visible to human eye. I
My aim is if i can train my net on that complex pattern and if my net can then catch that pattern on other products then i know i have made another good model..:clap:
Thx again for reply. I have pasting one of my strategies I made from NST. I already use the concept of normalisation for a robust system. The system I have attached is not the greatest but gives me some confidence that I have the ability to create winning systems.
I am a trader myself so i know what a net to output and if it doesnt outputwhat i want then something is wrong. For now I want my net to predict complex patterns which are visible to human eye. I
My aim is if i can train my net on that complex pattern and if my net can then catch that pattern on other products then i know i have made another good model..:clap:
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
arry u said that u trained net using eurusd data and used on xau data, how can i do this using neuroshell day trader? is it possible
I have trained a net on eurusd but i want to use it on gbpusd.. i want the information and pattern learned during eurusd to be directly implied on gbpusd as out of sample data, is this possible ( shall i saveeurusd as template and load on gbpusd and select match chart?
Hi Supremegizmo,
Congratulation for your good model, can you explain the detail what are input and output you are using for your net? Based on this information I may help you to identify whether you can implement your trained net can be applied in other pairs or not?
As far I know, If you are working with NSDT, you may save your trading template and use it for other pair, otherwise you can build a trading strategy where all pairs are loaded all prior you make optimization then it will be optimized for all pairs. If you are an eligible user for NSDT (have license), you can freely posting your question to Ward or open/download whole bunches from their support website (examples, indicators T&C magazine, etc).
So far I implement this way on NS2 and MT4, and I may verify to implement it on NSDT.
Congratulation for your good model, can you explain the detail what are input and output you are using for your net? Based on this information I may help you to identify whether you can implement your trained net can be applied in other pairs or not?
As far I know, If you are working with NSDT, you may save your trading template and use it for other pair, otherwise you can build a trading strategy where all pairs are loaded all prior you make optimization then it will be optimized for all pairs. If you are an eligible user for NSDT (have license), you can freely posting your question to Ward or open/download whole bunches from their support website (examples, indicators T&C magazine, etc).
So far I implement this way on NS2 and MT4, and I may verify to implement it on NSDT.
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
Arry bos i use a combination of %change in high low in order predict patterns like breakouts etc etc and output is 4% change in open
Today i try to load the template using eurjpy but the trades and patterns were not detected even though the same pattern where there in eurjpy .Instead i think it trained the net again.
It would be good if the net i trained on gbpusd is remebered exactly and the same memory is used for eurjpy.
Also i think my results are not the best they ok. I have better results generated by my advanced neural network software (my brain):cheesy:
In short I am supa smart in manual trading but with this nn i am helpless . I f i can put my manual strategy in NN then i know i am going to deliver some kick ass strategy
Hi Supremegizmo,
I have an idea how to check them, refer to your PM, we can have more discussion, one of my secret weapon :cheesy:
The are some weakness using our brain especially if our brain are not trained for any hidden issue..that's why most of trader are looser and most of the case the emotion driving all instead...:smart:
I have an advice, to check your manual trading consistency, you should prove your manual trading result at least for 3 months, on real account (not demo), shown it in myfxbook..then everyone can make conclusion.
See your PM for further discussion.
I have an idea how to check them, refer to your PM, we can have more discussion, one of my secret weapon :cheesy:
The are some weakness using our brain especially if our brain are not trained for any hidden issue..that's why most of trader are looser and most of the case the emotion driving all instead...:smart:
I have an advice, to check your manual trading consistency, you should prove your manual trading result at least for 3 months, on real account (not demo), shown it in myfxbook..then everyone can make conclusion.
See your PM for further discussion.
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
Hi Supremegizmo,
I have an idea how to check them, refer to your PM, we can have more discussion, one of my secret weapon :cheesy:
The are some weakness using our brain especially if our brain are not trained for any hidden issue..that's why most of trader are looser and most of the case the emotion driving all instead...:smart:
I have an advice, to check your manual trading consistency, you should prove your manual trading result at least for 3 months, on real account (not demo), shown it in myfxbook..then everyone can make conclusion.
See your PM for further discussion.
Arry i dont have fx book , but i will send my portfolio valuations in pdf extracted from my broker accounts
Amit,
My current progress, still developing dll file from c code created by NS2 using visual studio(not succeed yet).
Currently I just made indicator based on trained net from chaoshunter and aitrilogy..
My current progress, still developing dll file from c code created by NS2 using visual studio(not succeed yet
).Currently I just made indicator based on trained net from chaoshunter and aitrilogy..
Attachments
arryex; said:Amit,
My current progress, still developing dll file from c code created by NS2 using visual studio(not succeed yet
Currently I just made indicator based on trained net from chaoshunter and aitrilogy..
Arry,
What your problem with dll or vs, i can help. I worked in that environment so many years.
