Neural Networks - page 18

 

Neuro forex trend strength predictor2

Hi Guys,

What a nightmare, hard drive fried on my laptop so i went out this morning and bought a new one...fine, but all my software was on the old laptop, spent all sunday reinstalling software.

One thing i miss the most is the neuro forex trend predictor 2, well i found a download somewhere earlier on, it will do, but the one i wanted was from another forum.

Hopefully someone here might have seen it, whoever posted it on that forum added a snake to the NFTSP, i found it very accurate and enjoyed using it. sadly the other site, i think has been taken down or is not working today.

So heres my question, did anyone d/l from another forum the NFTSP2 (WITH SNAKE), if so, do you still have it, if so, can you send it to me please, it would be most helpful.

thx guys

 

Neural Networks FOREX Trading

Using Neural Networks to create FOREX Trading Strategy (About this text)

In this free online tutorial you will find the "full cycle" of using neural networks (Cortex Neural Networks Software) for Forex trading (or stock market trading, the idea is the same).

You will learn how to choose inputs for the artificial neural networks, and how to decide what to use as the output.

You will find an example of a ready to use script that allows to perform neural networks optimization of both the structure of Neural Network (number of neurons) and the forex trading system (stop loss etc.)

Finally (the part that is not present in most tutorials), you will learn what to do next. After all, Cortex Neural Networks Software cannot do real time trading, you need to use something like Trade Station, MetaQuotes or MetaTrader. How to port the FOREX trading system from Cortex to your favorite trading platform? Do you have to deal with DLLs, ActiveX controls and low-level programming? The answer is NO.

Cortex Neural Networks Software comes with the easy to use feature that allows you to easily port the resulting (trained) Neural Network to the scripting language of your trading platform. No DLLs, DDE, ActiveX or any other low-level solutions - everything is plain and simple.

Important note: this is NOT a "how to trade" tutorial. Instead, it tells you how to use Cortex Neural Networks Software, but you still need to invent your own trading system. The one we use here is barely a starting point, and shouldn't be used as a forex trading strategy "as is". The idea of this text is to teach you to create NN-based trading systems and to port them to the trading platform of your choice. The example is, however, ovesimplified, and can only be used as the illustration of trading principles. Same way, the MACD trading system, that can be found in many tutorials, is not working well anymore (as markets have changed), but still is a good example of using indicators for mechanical trading.

In two words: do your own analysis.

Another important note: the tutorial is using examples, lots of them. To make your life easier, I have included them all, not just fragments. However it makes the text much longer. Also, I am going from the very first, clumsy, forex trading system, to more advanced, every time explaining what had been improved and why. Be patient, or jump directly to the section you need.

Final important note: the code is not something carved in stone, it could change while this text was written. The final versions of script files are included in Cortex archive.

The link to this tutorial (with source code included) : Using Neural Networks to perform FOREX Technical Analysis. Creating FOREX Trading Strategy. Profitable FOREX Trading System Software.

 

Only now found this post. Great link !

Thanks

 

Machine Learning

If anyone is interested in developing machine learning based strategies, check out www.deep-thought.co.

Currently supports

  1. Support Vector Machines.
  2. Gradient Boosted Trees.
  3. Random Forests.
  4. Extremely Randomised Trees.
  5. Multi-layer Perceptron, aka Neural Network.
  6. Ensembles: Combine the forecasts of any number of predictors.
  7. Constant retraining, always adapting to the market.

Please note, this is a tool to develop your own strategies and systems, not a pre-canned trading strategy.

Also included are two MT4 EAs, with source, to trade the signals or combine with any other system you may have.

 

Thank you kiwifxguy,

What are the retraining times like if we use constant retraining? If we export indicator csv data, can it be used as an input and normalized within deep -thought?

Looks good overall although I just skimmed though the manual. I will have to look at it in more detail a little later.

kind regards,

Alex

 
hughesfleming:
Thank you kiwifxguy,

What are the retraining times like if we use constant retraining? If we export indicator csv data, can it be used as an input and normalized within deep -thought?

Looks good overall although I just skimmed though the manual. I will have to look at it in more detail a little later.

kind regards,

Alex

Hi Alex,

The retraining depends on how many predictors in the ensemble and the speed of the PC. As a guide I'm using an ensemble of around 80 SVMs and retraining takes around 5 minutes on an elderly single core laptop.

The sequence of events is that when a candle completes, the forecasting is done and orders/trades placed etc. After all that the retraining starts. I'm using a 4 hour time frame so the newly trained models aren't needed until the end of the 4 hour candle, so it finishes in plenty of time.

As for csv indicators from MT4, that isn't supported yet. The main problem is getting historical and reliable values. I've found MT4 a bit flakey in this area. Was going to look at providing the ability to code indicators directly in DeepThought using Python and/or R in the next release - there's a Python wrapper for TA-lib (TA-Lib : Technical Analysis Library - Home) which has zillions of indicators,

Cheers

Jonathan

 

Hi Jonathan,

Have a look here. You might find the exporter interesting. This is how I am getting indicator data into Rapidminer. I have been using this for a couple years now.

https://www.mql5.com/en/forum/181252

Your retraining times are interesting. To give you an example, using MySVM from Stefan Ruping training on daily data to forecast the next days high or low was taking roughly 60 minutes using a start date of Jan 2010 with five inputs running two threads.

I am now using a completely different method for estimating the next days range which you can see here. I am using three take profit targets with the final target being the estimated target. This is a real time snapshot of EURUSD this morning. Still I think I could get even better accuracy by combining both techniques.

Regards,

Alex

I will contact you via private mail.

Files:
eurusdh1.png  34 kb
 

Hi Alex,

The exporter looks interesting, I might try something similar.

The training set I'm using (used by all SVMs in the ensemble) has 116 features, and 1000 training samples. Some raw price changes, moving average differences, time of day and day of week.

DeepThought uses libSVM but I thought it should generally have similar performance to MySVM in time and accruary, although I've made small mods to the source to make it use all cores. Increasing the number of training samples beyond this doesn't seem to make much difference to the results but does increase training time,

Cheers

Jonathan

 

NNTREND v.2 expert advisior

Found on another forum an updated version if the old is not working, updated plow, but choen many settings, I have not figured out.

_www.sendspace.com / file / wuzhyh

instruction in archive.

Files:
test.jpg  133 kb
 

No idea if it is nn indicator - it is interesting repro_neuro_trade.ex4

Files:
rnn.gif  29 kb
Reason: