Using neural networks in trading - page 15

New comment
 
Patience...
 
grell:
And if the output is not forecast, then how?
Where did I say about prediction? Learning is about minimizing the output error. That's the error we need to look at to see if the network has learned anything.
 
TheXpert:
Where did I say anything about prediction? Learning is about minimising the output error. That's the error to see if the network has learned anything.


Wrong again. How will I measure the error if the output is an erroneous entry into a trade? You can't. Because the network must have an example and it does not.
 
grell:
Missed again. How do I measure the error if the output is an erroneous entry into the trade? You can't. Because the network must have an example and it doesn't.
You're digging somewhere in your cockroaches now, you don't want to get into it, don't get into it.
 

This is a 5 year sample, a learning period. Drawdown 934.87, net profit 1396.06. This is without one neuron.

 

Here's a normally trained network on a training sample --


 
TheXpert:
You're digging somewhere in your cockroaches now, you don't want to get into it, don't get into it.


The objectives of the network are different.
 
TheXpert:

Here's a normally trained network on a training sample --



What period is this?
 
grell:
What period is that?
I don't remember. It must have been a year.
 
TheXpert:
I don't remember. It must be a year.


The result is good, it looks like a fit, but it's more the fault of the sample than the network. Next screen will be training with additional neuron, but it's a year. First screenshot 1.1.2008-1.1.2013. Second screenshot 1.1.2008-1.1.2009. Still learning:)
1...8910111213141516171819202122...40
New comment