Machine Learning

kiwifxguy

Junior member
Messages
17
Likes
0
If anyone is interested in developing machine learning based strategies, check out www.deep-thought.co.

Currently supports

  • Support Vector Machines.
  • Gradient Boosted Trees.
  • Random Forests.
  • Extremely Randomised Trees.
  • Multi-layer Perceptron, aka Neural Network.
  • Ensembles: Combine the forecasts of any number of predictors.
  • 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.
 
Sounds like one needs a PhD in CS to use such software:)
Actually a PhD in any quantitative discipline should be enough ... :)

But point taken, it can deem a little daunting at first but then most things are. The short term pain of learning should not dissuade from trying new directions in trading. The interest in machine learning in trading is increasing with prop shops in particular taking a keen interest, so I believe it's worth looking into.
 
Tutorials

I've added a couple of tutorials to the website and some example configurations to the downloads section.

Also added the ability to use MT4 indicators and fixed the DLL so it works with all the breaking changes that build 600 introduced.
 
Hi,
In your website it says ''GA can optimise on PnL, Sharpe Ratio, Accuracy'' Do you mean win/loss % by accuracy?

1-Can you also add Maximum Adverse Excursion Percent filter feature http://mysimplequant.blogspot.com.tr/p/trading-system-performance-indicators.html as fitness function choice for the GA option? So the GA can optimise on minimum maximum adverse excursion percent?For example if the signal is below the certain percent threshold the GA will filter out the signal?

2- Will you add self training NN's, online learning algos http://leon.bottou.org/projects/lasvm ,http://onlinesvr.altervista.org/ http://stats.stackexchange.com/ques...ining-data-to-existing-svm-models/51989#51989 to your software?Thanks.

By the way I am interested in you software and I am studying your software manual and the software features.
 
Last edited:
Hi,
In your website it says ''GA can optimise on PnL, Sharpe Ratio, Accuracy'' Do you mean win/loss % by accuracy?

1-Can you also add Maximum Adverse Excursion Percent filter feature http://mysimplequant.blogspot.com.tr/p/trading-system-performance-indicators.html as fitness function choice for the GA option? So the GA can optimise on minimum maximum adverse excursion percent?For example if the signal is below the certain percent threshold the GA will filter out the signal?

2- Will you add self training NN's, online learning algos http://leon.bottou.org/projects/lasvm ,http://onlinesvr.altervista.org/ http://stats.stackexchange.com/ques...ining-data-to-existing-svm-models/51989#51989 to your software?Thanks.

By the way I am interested in you software and I am studying your software manual and the software features.

Hi Tovim,

The accuracy (aka precision in stats terminology) is wins/(wins + losses). I'm currently implementing Python scripting so you should be able to create your own GA objective function(s). The python scripting will also be able to be used to create features, custom predictors and logic for enembles.

I've had a quick look at the source of lasvm and it looks like it shouldn't be too difficult to add in. DeepThought currently uses libSVM, libLinear and the machine learning library from OpenMP. It was designed with an architecture to simplify adding of predictors. I'm also interested in Deep Learning/RBM and will add that in sometime.
 
Hi ,

Since MQL4 platform becoming increasingly unprofessional and unreliable platform for automated trading , do you plan to look for free broker API solutions?

Also this ML algo looks interesting: Extreme Learning Machines trains hundreds of times faster than SVM and RMSE for both ELM and SVR are almost the same.See the comparison : http://www.ntu.edu.sg/home/egbhuang/ online sequential ELM http://www.ntu.edu.sg/home/egbhuang/elm_codes.html
 
Last edited:
DeepThought now has Python scripting for feature generation, target/trigger, predictors and buy/sell signal generation logic.

The next major release will include a GUI as it's currently command-line driven which I guess can be a little off-putting to those not used to command-line interfaces.

www.deep-thought.co
 
are you offering a trial of this ??

Krzysztof

There's no trial at the moment, but a trial version will be available when the GUI is released. I'm the meantime if you buy it and change your mind I'll do a refund.

When the GUI is release I'm thinking about opensourcing some of the commandline components.
 
There's no trial at the moment, but a trial version will be available when the GUI is released. I'm the meantime if you buy it and change your mind I'll do a refund.

When the GUI is release I'm thinking about opensourcing some of the commandline components.

How do you deal with data snooping?

In your website there is no name, no address, no telephone.
 
For anybody interested in learning about Machine Learning there is a free course from Stanford University on Coursera starting on 16th July. It is taught by Andrew Ng, one of the leading lights in ML and a good teacher.
https://www.coursera.org/course/ml
I took the course in 2012 and gained score of 99.6%. It was hard work but enjoyable and satisfying.
 
Hi Solas,

How do you deal with data snooping?

Not sure what you mean by data snooping. If you mean can a backtest 'see' into the future, then it doesn't. Candles are pushed through in the same way as they would appear from the live market. This means that training examples have to be 'cached' until a target has been reached before the training example can be labelled.

In your website there is no name, no address, no telephone.
My name is in the FAQ in the question 'Who is behind DeepThought'.
 
For anybody interested in learning about Machine Learning there is a free course from Stanford University on Coursera starting on 16th July. It is taught by Andrew Ng, one of the leading lights in ML and a good teacher.
https://www.coursera.org/course/ml
I took the course in 2012 and gained score of 99.6%. It was hard work but enjoyable and satisfying.

Well worth doing this course. You can also see the lectures on Youtube:
www.youtube.com/watch?v=UzxYlbK2c7E&list=PLJ_CMbwA6bT-n1W0mgOlYwccZ-j6gBXqE, or search for 'machine learning' in iTunesU.
 
There's no trial at the moment, but a trial version will be available when the GUI is released. I'm the meantime if you buy it and change your mind I'll do a refund.

When the GUI is release I'm thinking about opensourcing some of the commandline components.

To contradict myself, there is now a 60 day trial of the current commandline version. Everything's in the trial except the dll required for live trading. Available in the downloads section: http://www.deep-thought.co/dt-downloads.
 
Top