estimates fail when the name of the input timeset isn´t the the name of the output timeset

[imendez]

If the name of the inputs timesets is not the same that the name of the output
timeset, Estimate function fails, although the timeset is really the same.

The code is:

/
Date IniEst = y2004m01d01;
Date EndEst = y2004m10d01;

TimeSet TmsY = Diario;
TimeSet TmsX = Diario;
Serie y = SubSer(Trend(IniEst, TmsY)+Gaussian(0, 0.2, TmsY), IniEst, EndEst);
Serie x = SubSer(Trend(IniEst, TmsX), IniEst, EndEst);

Set ModDef = ModelDef
(

y,
1,
0,
1,
0,
1,
SetOfPolyn(1, 1),
SetOfPolyn(1, 1),
SetOfSet(InputDef(0.1, x)),
If(ObjectExist("Set", "AllLinnear"), AllLinnear, AllLinear)

);

Set EstLineal = LinearRegression(y, SetOfSerie(x));
Set Estimacion = Estimate(ModDef, IniEst, EndEst);
/

and the error retruned is:

/
ERROR: Fechados incompatibles en las series output y/o input.No se obtuvieron los
datos de las series input.
/

HowEver, if I use "Diario" instead of its copies "TmsX" and TmsY", no errors ocurr.
Notice that LinearRegression works perfectly.

This code works until the 20040929 version and start failing at 20041006 version.

[Víctor de Buen Remiro]

There is no analytical way to know if two dating are the same or not, because a
dating is a virtual infinite object. Exhaustive methods could be applied to finite
time intervals but it should implies an expensive cost of processing time, because
it should affect to any time series operation.

Until bug 161 (see ​http://bugs.tol-project.org/cgi-bin/bugzilla/show_bug.cgi?
id=161) was resolved, Estimate function didn't worry by this situation and ask for
data to all input and output between two dates. If dating was different results was
unpredictable and very difficult to debug it.

To avoid this problem you simply must to use the same dating in all input and
output. Certainly it's neccessary that all related functions as LinearRegression
works in the same way.

[Víctor de Buen Remiro]

As there are very problems similar to this one for some TOL analyst, I've changed
Estimate behavior in this sense. Now you can use Estimate like any other time
series fucntion: the only one thing required is all related dating in output and input
series are compatible between them selves, it's to say, all them must to have the
same dates in calculus interval of dates.