Need to know if Neural Networks can achieve my goal
in [Matlab]
|
From: Greg Heath on 23 Jun 2010 20:02 On Jun 23, 6:35 am, "Franco " <fjmo...(a)gmail.com> wrote: > Hey there, > > I want to start learning neural networks to help me with trading. But before I do that I want to confirm that this is the method I need to learn. > > For example, see below for the images: > > http://img101.imageshack.us/i/patternr.gif/http://img227.imageshack.us/i/pattern2.gif/ > > Both images contain a pattern also known as a whipsaw or a sideways pattern. The price is not really going anywhere it is just oscillating up and down. Sometimes tis happens with an overall upwards on downwards trend, other times there is no trend is just oscillates up and down. Now I need to detect those anomalies mathematically. > > Is this possible with neural network pattern recognition? It looks like a time-delay NN like the function newfftd could work. Just create a target output file that contains a 1 if the corresponding input point is in a whipsaw pattern and a zero otherwise. Of course you have to find a sufficient amount of data for training, validation and testing.. Probably the most important parameter (besides the number of hidden nodes to have) is how much leading data does the net need before it can confidently label a point as a member of a whipsaw interval. Apriori knowledge can get you a ball park intial estimate. However, trial and error is king. Good Luck, Greg
From: Franco on 24 Jun 2010 02:39
Greg Heath <heath(a)alumni.brown.edu> wrote in message <4a8e79ef-d7f2-4a94-9b7b-21adf9144a23(a)g19g2000yqc.googlegroups.com>... > On Jun 23, 6:35 am, "Franco " <fjmo...(a)gmail.com> wrote: > > Hey there, > > > > I want to start learning neural networks to help me with trading. But before I do that I want to confirm that this is the method I need to learn. > > > > For example, see below for the images: > > > > http://img101.imageshack.us/i/patternr.gif/http://img227.imageshack.us/i/pattern2.gif/ > > > > Both images contain a pattern also known as a whipsaw or a sideways pattern. The price is not really going anywhere it is just oscillating up and down. Sometimes tis happens with an overall upwards on downwards trend, other times there is no trend is just oscillates up and down. Now I need to detect those anomalies mathematically. > > > > Is this possible with neural network pattern recognition? > > It looks like a time-delay NN like the function newfftd could work. > Just create a target output file that contains a 1 if the > corresponding > input point is in a whipsaw pattern and a zero otherwise. Of course > you have to find a sufficient amount of data for training, validation > and testing.. Probably the most important parameter (besides > the number of hidden nodes to have) is how much leading data > does the net need before it can confidently label a point as a > member of a whipsaw interval. Apriori knowledge can get you > a ball park intial estimate. However, trial and error is king. > > Good Luck, > > Greg Thank you very much for your help. I have read that neural networks has great potential and it is up to the trainer to make it work. I will do my best. Regards Franco |