Hi everybody,
I hope you've had a happy Christmas, where that applies!
This is just to inform you that the contributed function package tsfcst
is going to be updated shortly. Why this message? Because the package is
not really maintained anymore, so the gretl team is just doing some
housekeeping on it. This means that the upcoming version 0.6 will
require a fairly recent gretl version, but more importantly the
functions NMmaxC, NMminC, and hinv are going to be removed from the package.
The mentioned functions weren't needed for the main functionality, so
the rest of the package is not affected. However, there may be a small
possibility that someone was using those specific functions, although it
seems relatively unlikely. For those of you let me just paste the hansl
code of those functions directly at the bottom of this message. (Notice
that there seems to be a typo in one line; we haven't checked how
relevant that line is for typical usage.)
Finally, to give you some context, the package apparently implements
some methods that can also be found in an R package by Hyndman. For your
convenience here's a snippet from the package's help text, but be sure
to check out the entire help in the package for more details (until it
is moderated and really published, the updated version is in staging
here: http://gretl.sourceforge.net/staging_fnfiles/tsfcst.gfn):
<tsfcst-help>
...
PACKAGE DESCRIPTION
===================
The "main" forecasting function is stheta(), which implements the
theta method, including a version for seasonal data; expsmpars()
outputs auto-selected alpha and y_0 to use for exponential smoothing.
...
</tsfcst-help>
OK, so much for that, enjoy the rest of this ugly year!
cheers
sven
--------------------
# Old functions from Oleh's package tsfcst <= v0.53
# which didn't really have to do with the rest of the package.
# Apparently does constrained (Nelder-Mead) optimization
# by mapping the function from 0-1 domain to infinite domain
# via the hyperbolic tangent inverse or something.
function matrix NMmaxC (string fu "quoted expression to minimize",
matrix los "left bounds",
matrix his "right bounds",
matrix z[null] "other variables",
matrices M[null] "yet other variables",
list L[null] "further variables as series",
matrix x[null] "init. val. of optimizing variable")
# inside fu the optimizing variable is 'x'
# M is needed when one has many matrices of different shape
if !exists(x) # was isnull()
matrix x = start_mt(los,his)
else
matrix x = x
endif
fu0 = fu
fu = smartstrsub(fu,"x","hinv(x,los,his)")
n = nelem(los)
cns = n == 1 ? "fmax x" : "fmax"
if n == 1
xi = "x[1]"
txtxi = "hinv_sc(x[1],los[1],his[1)" # must be a typo here
else
loop i = 1..n -q
xi = sprintf("x[%d]",i)
txtxi = sprintf("hinv_sc(x[%d],los[%d],his[%d])",i,i,i)
fu = strsub(fu,xi,txtxi)
cns = cns~" "~xi
endloop
endif
fu = "("~fu~")"
nm = NMmax(x,@fu)
x = hinv(x,los,his)
fmi = @fu0
matrix ret = fmi~x
cnameset(ret,cns)
return ret
end function
function matrix NMminC (string fu "quoted expression to minimize",
matrix los "left bounds",
matrix his "right bounds",
matrix z[null] "other variables",
matrices M[null] "yet other variables",
list L[null] "further variables as series",
matrix x[null] "start values of 'wrt. to'
variable")
# inside fu the optimizing variable is 'x'
# M is needed when one has many matrices of different shape
if !exists(x) # was isnull()
matrix x = start_mt(los,his)
else
matrix x = x
endif
fu0 = fu
fu = smartstrsub(fu,"x","hinv(x,los,his)")
n = nelem(los)
cns = n == 1 ? "fmin x" : "fmin"
if n == 1
xi = "x[1]"
txtxi = "hinv_sc(x[1],los[1],his[1)" # again a typo apparently
else
loop i = 1..n -q
xi = sprintf("x[%d]",i)
txtxi = sprintf("hinv_sc(x[%d],los[%d],his[%d])",i,i,i)
fu = strsub(fu,xi,txtxi)
cns = cns~" "~xi
endloop
endif
fu = "("~fu~")"
nm = NMmin(x,@fu)
x = hinv(x,los,his)
fmi = @fu0
matrix ret = fmi~x
cnameset(ret,cns)
return ret
end function
function matrix hinv (matrix x "arguments of objective function",
matrix lo "lower bounds in restrictions",
matrix hi "upper bounds in restrictions")
# box -> R^n transformation
nl = nelem(lo)
matrix hix = x
loop i = 1..nl
hix[i] = hinv_sc(x[i],lo[i],hi[i])
endloop
return hix
end function
function scalar hinv_sc (scalar x,
scalar lo,
scalar hi)
if hi <= lo
funcerr "'hi' cannot be less or equal to 'lo'."
endif
if (lo > -$huge) && (hi < $huge)
return (lo+hi) / 2 + (hi-lo) / 2 * tanh(x-(lo+hi)/2)
elif (lo>-$huge) && (hi>=$huge)
return lo + exp(x-lo)
elif (lo<=-$huge) && (hi<$huge)
return hi - exp(hi-x)
else
return x
endif
end function
Hi, *Gretl and Panel Data Analysis Experts.*
Using Gretl, how to perform Fixed-Effect and Random-Effects models for
panel data with a Dichotomous DV?
Thank you for your inputs.
Happy holidays.
Lema
Dear Profs,
I hope this email finds you well.
Please, I've got to realize recently that the impulse response function in
the SVAR and the SVECM is no longer working. It keeps showing error
messages. Only the historical decomposition is functioning.
I humbly request that the authorities in the house help look at what is
wrong. Maybe the SVAR/SVECM routine is no longer compatible with the newly
released version of the grelt.
Thanks.
Dear Prof.
I have tried updating the SVAR add-on package to version 1.96 on my PC, yet
the problem still persists. In fact, the bugs have compounded; the help pdf
file could not even be opened, even from another PC.
Find below the error messages:
C:\Users imme\AppData\Roaming\gretl\irf768.gp: Invalid argument
*** error in function IRFgrph, line 34
> outfile "C:\Users imme\AppData\Roaming\gretl\irf768.gp"
called by function IRFsave
*** error within loop in function IRFsave
> tmpfile = IRFgrph(IRFmat, snum, vnum, scale, sn, yn, keypos, null, &boot,
bc, whichdraw)
called by function IRFplot
called by function GUI_plot
C:\Users imme\AppData\Roaming\gretl\fevd9937.gp: Invalid argument
*** error in function FEVDgrph, line 22
> outfile "C:\Users imme\AppData\Roaming\gretl\fevd9937.gp"
called by function FEVDsave
*** error within loop in function FEVDsave
> string tmpfile = FEVDgrph(Fmat, vnum, obj.Ynames[vnum], obj.snames,
keypos, titletail)
called by function FEVDplot
called by function GUI_plot
C:\Users\timme\AppData\Roaming\gretl\.mysession\functions.pdf: No such file
or directory
Hi again, let me put this information about new features in the
development (= not yet released) version into this separate message.
(And no, there's no schedule yet for when the next version will be
released.)
I'm not going to repeat the entire changelog which I linked to in the
previous message; this is just about the stuff that might be most
interesting for the end user:
- A hansl (scripting) function distance(): "evaluate the distance
between points on various metrics such as Euclidean, Manhattan, Hamming,
Chebyshev, Cosine and Mahalanobis"
- Also for scripting: you can now use the keyword "end" for indexing
into multi-element objects, to denote the last entry or item. I'm not
sure if this is already documented, by the way, but basically "end" can
stand in for rows(x), cols(x), nelem(x), depending on the context.
- Panel data time dimension: This handling has been much improved. For
example, creating a new panel dataset in the GUI is now much easier, you
can directly enter the wanted n and T dimensions. And restricting the
sample in the GUI is now also possible in the time dimension.
Again, this only concerns the next version (and current snapshots)!
thanks
sven
Dear all,
many thanks for accepting me in this thread.
I have been a student user of Grelt and now trying to grab it again for PCA modelling in my current work.
I tried to replicate the score calculation as provided by Gretl (PC1, PC2 etc vectors that you can store and save after running the PCA command) but i cannot seem to derive those as simply:
PC Factors = Eigenvectors * standardized original variables (x - mean).
Where am i making a mistake?
Thanks for all your wonderful contribution,
Best
Francesco
Hi,
can I define a time-varying constant in the transition equation as
B.stconst ={mu*B.signt}
where B.signt is a scalar updated by B.timevar_call "update function"?
It is useful to estimate mu, but I'm not sure B.stconst is updated at
each time!
Thanks,
Paolo
