Re: [Gretl-devel] [Gretl-users] GUI enhancements
sorry i was using french words.
for example :
you have 250 values of a unique curve.
if You want to forecast the 251th and would like to know the real
accuracy of the 100 pasts value.
So you need to put 150 values and do the forecast of the 151th then
151 values and the of the 152th...and so on...
if you go 10 exact values on 100 tries, you have 10% probabilities to
have the real values of the 251th (and more you take example more you
have a forecast probabilities).
Then you change parameters to see you got more probabilities with the forecast.
My usage is to forecast if a value will be greater(or equal) or lower
of the 250th (i test every 100 past values to know the probabilities
of the forecast)
and i change parameters to optimize the probabilities.
I used it before with learning algorithm (like kernel machine, or perceptron).
2010/6/17 Allin Cottrell <cottrell(a)wfu.edu>:
On Thu, 17 Jun 2010, denis joubert wrote:
> I have a request too :-) (or a question i don't know if it's
> already there) Is it possible to have a validation process of
> forecasts. I mean using past values to verify forecast (without
> using it to the "learning" or modeling process) ? because it's
> what i implemented under cygwin with libgretl... and more than
> all i use my validation process in order to have more accurate
> parameters of model (number of lags is one the parameters i try
> to optimize)
Sorry, not sure exactly what you mean by "validation" or
"verification" of forecasts. Can you explain a bit more?
Allin
_______________________________________________
Gretl-devel mailing list
Gretl-devel(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-devel