Testing real-time forecasting systems - page 31

 

while in the zone... (middle of the fourth day)


 
grasn >> :

to gpwr

I have several models in operation at the moment. For this one I have not gathered complete statistics, it is quite expensive computationally, but the "spot test" on relatively large areas (but not on the full history) shows good results. I am going to "add" NS (some ideas have appeared), and thoroughly test on history, but this time evaluating by deals, but not by errors. If I look at the statistics I will use probability nets, but what kind of nets I will need to use carefully.

For now I shall have to use them.


dimasik 03.09.2009 11:45

Dear gpwr, What parameters do you set for your indicator? I will use it as an indicator and may use it for determination of my values, thank you.

2*sigma = 0.1

The number of past patterns is limited to 500. If you take the whole history then the prediction error is reduced. It turns out that very old patterns are only detrimental.

There are some nuances of what data is passed to the network inputs. I don't use naked prices. There are also some subtleties in the network code, i.e. formulas used are slightly different than those in Specht's article

Previous forecast


Here is the new GRNN forecast.


 

I will not recalculate anything yet, I wonder how much the forecast will deviate from the fact by the end of the week. Right now the situation is as follows:


 
GPWR, when trying to put an updated version of Microsoft Visual C++ indicator on the chart, it gives error "Runtime error".... How do I fix it?
 
dimasik >> :
GPWR, when trying to put an updated version of Microsoft Visual C++ indicator on the chart, it gives error "Runtime error"..... how do you fix it?

which indicator? BPNN Price Predictor? I am not using it yet. I've actually gotten cold to direct-distribution networks. I have a new hobby now: probabilistic networks and input data preparation (compression). I can't say anything about the error. Sorry I can't help.

 
gpwr >> :

which indicator? BPNN Price Predictor? I don't use it yet. Actually, I've cooled down to direct propagation networks. I have a new hobby now: probabilistic networks and input data preparation (compression). I can't say anything about the error. Sorry I can't help.

similarly :o)

PS: The taste of NS is coming back :o)

 
gpwr писал(а) >>

What indicator? BPNN Price Predictor? I don't use it yet. Actually, I've cooled down to direct propagation networks. I have a new hobby now: probabilistic networks and input data preparation (compression). I can't say anything about the error. Sorry for not being able to help.

Is there a GRNN indicator freely available? I would like to test it on other instruments. The results on eu are impressive.

 

All in all the day went well, forecast for Friday:

gpwr, preferably along with the forecast for the future to show how the previous forecast worked out

 
neoclassic >> :

gpwr, it is advisable to show how the previous prediction worked along with the future prediction

Added an old prediction to my previous post.

 
Colleagues, does your forecasting quality depend on scale (timeframe) ?
Reason: