|
PureBytes Links
Trading Reference Links
|
At 01.39 13/02/03 +0000, you wrote:
>I just love the "idea" of NN's except for the fact that what you wind
>up with is usually a black or at best a gray box
It may sounds strange to you, but I started using NN just for the
"black-box" idea:
what I tought is: "there may be a 'strange' use of data that can build a
good trading system, but this strange combination may be very
controintuitive and may be achieved only with a long and difficult
computation".
May be interesting to understand how a black blox works but, for me, it's
much more important that black box works correctly and gives me a very good
trading system. I've never seen performances like the ones achieved by my
NN systems.
>and that I've yet to
>see any NN model that did anything worth talking about on out of
>sample data
when you train a neural network, you work with, at least, two set of data:
train data and cross validation data (there are, also, test data).
While you train the network with train data, at the same time you test it
on cross validation data i.e. NN is trained with a set of data and is
tested on data never seen by the network. The performance, and stop
criterion, are based on the network performance achieved with cross
validation data i.e. data never seen before.
>As far as Neurosolutions offerings I can't the demo of their product
>to crunch numbers even when supplied with fairly simple inputs. It
>just hangs.
Not to me: NN works correctly to me. It may be problems with data (or file)
format.
regards
Buttha
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
|