I started to play with MemBrain this week. I have used other NN software (like faNN and Joone) but MemBrain seems quite interesting and much easier to use. One thing I still cannot figure out is how to use time series in it, which seems to me to be very important for any attempt to forecast. Do you or anyone know how to do that in MemBrain or can you tell me if it is even possible?
Would you mind sending me source code for the wrapper? My email is my username at gmail.com
every new pattern you create can be a new point in time. In the teacher menu you can set the lesson pattern selection to Ordered. But please don't try to forcast with a BP NN. It will cost you a lot of time and will give you a lot of frustration. It just won't work, take my word for it. Better invest on reinforment learning (RL) / genetic algorithms (GA) / etc. Currently I'm working with RL RNN's which are great for adaptive EA's.
- Sutton & Barto
Are these algorithms you refer to available in Membrain or are you using yet another NN application?
I am not in fact trying to forecast the prices. My idea right now is to feed it the signals via a CSV file with prices and signal (+1 buy, -1 sell and 0 do nothing) and train the NN so it can suggest buy and sell signals with current data. Do you think that is possible with Membrain? So far I seem to get extremely high net error but since there are so many parameters in Membrain it can perhaps be caused by some obscure one that I don't completely understand.
I will look into those references. I had a look at a couple and they seem to be rather interesting reads. I am just not sure how to implement that in Membrain.
Thank you for all your help!
No they are not available in MB. Q & TD learning are alg's from Reinforcement Learning. Though they can be used with NN's, but it takes a different way how errors are derived; amongst other things. I have tried the idea you suggested and couldn't get it to work. After several weeks I gave up the approach and switched to RL. Main problem is after training the net it will allways generate the same "reply". But market conditions change over time, so it gets out of sync (like technical indicators). That's why a adaptive system will have have higher probability of success.
At this point I'm not using Membrain. I wrote a BP NN in metatrader and use it for RL (the QL alg)
I wrote a wrapper for membrain... http://www.membrain-nn.de/main_en.htm
here the files for those how like to experiment with it. I haven't fully tested it yet. So any feedback is welcome
I can download the MQ4 and MQH files but I cannot download the wrapper dll. Has it been removed because of licencing reasons?
I have installed VS10 so if you could kindly post the (I assume C++) code of the wrapper it would may be even better. I am not very familar with the C++ header etc. stuff but I wrote some basic C++ dll for MQ4 use some time ago, so I could compile it on my own (and would also learn something). I have installed Membrain and take the first steps guided by a German book but I would like to see your sample code working, too, before acquiring a licence.
Kind regards, Jack Rabbit
I'm just experimenting with membrain and have severe problems to have the membraindll.dll have a so called "thinkstep" executed.
As you wrote this wrapper for Membrain once upon a time,
could I ask you something about the interaction between Membrain and MetaTrader4?
cany anyone prvide me the wrapper dll ?
Thanks in advance