4:50 a.m.
Hi,
I just stumbled over old hansl code by Jack that implements the
fractional power of a pd matrix (see at the end). The comment hints at
the plan to integrate this somewhere, but it never happened. Should it
go into the 'extra' addon?
thanks
sven
----
function matrix pdmatpower(const matrix A, scalar pow)
# Helper function by Jack Lucchetti to calculate
# a fractional power (e.g. root) of a pos-def matrix.
# Once this feature is in extra.gfn or in core gretl
# it can be retired.
matrix ret = {}
catch l = eigensym(A, &ret)
err = $error
if err
msg = sprintf("%d - %s", err, errmsg(err))
funcerr "@msg"
elif minc(l)<0
msg = sprintf("matrix %s is not pd", argname(A))
funcerr "@msg"
else
l = l .^ (pow/2)
ret = ret .* l'
ret = ret*ret'
endif
return ret
end function
7:46 a.m.
On Fri, 26 Feb 2021, Sven Schreiber wrote:
Hi,
I just stumbled over old hansl code by Jack that implements the
fractional power of a pd matrix (see at the end). The comment hints at
the plan to integrate this somewhere, but it never happened. Should it
go into the 'extra' addon?
It's now in git, under the name "psdpwr" (because in fact it applies to
positive SEMI-definite matrices). If you're ok with it I'll write the
documentation as soon as I get the chance to.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
8:07 a.m.
On Fri, 26 Feb 2021, Riccardo (Jack) Lucchetti wrote:
On Fri, 26 Feb 2021, Sven Schreiber wrote:
> Hi,
> I just stumbled over old hansl code by Jack that implements the
> fractional power of a pd matrix (see at the end). The comment hints at
> the plan to integrate this somewhere, but it never happened. Should it
> go into the 'extra' addon?
It's now in git, under the name "psdpwr" (because in fact it applies to
positive SEMI-definite matrices). If you're ok with it I'll write the
documentation as soon as I get the chance to.
One thought on this: we already have built-in support for raising a
matrix to a power (fractional or integer) as a whole, as opposed to
".^" for the element-wise case. See matrix_frac_pow() in
gretl_matrix.c
So wouldn't it make sense just to inflect that code to handle the
PSD case specially, rather than adding a special function?
Allin
9:02 a.m.
Am 26.02.2021 um 14:07 schrieb Allin Cottrell:
On Fri, 26 Feb 2021, Riccardo (Jack) Lucchetti wrote:
> On Fri, 26 Feb 2021, Sven Schreiber wrote:
>> I just stumbled over old hansl code by Jack that implements the
>> fractional power of a pd matrix (see at the end). The comment hints at
>> the plan to integrate this somewhere, but it never happened. Should it
>> go into the 'extra' addon?
>
> It's now in git, under the name "psdpwr" (because in fact it applies
> to positive SEMI-definite matrices). If you're ok with it I'll write
> the documentation as soon as I get the chance to.
One thought on this: we already have built-in support for raising a
matrix to a power (fractional or integer) as a whole, as opposed to
".^" for the element-wise case. See matrix_frac_pow() in gretl_matrix.c
So wouldn't it make sense just to inflect that code to handle the PSD
case specially, rather than adding a special function?
Actually, I just checked section 17.6 of the guide, where it says that
the matrix must be PSD if the power is fractional. (And must be
invertible --and thus PD-- if the power is negative.)
So maybe the whole issue is moot because it is already fixed and done?
Three follow-up questions on this:
- The power operation should be added to table 17.2 of the guide, not
only mentioned in the text of section 17.6.
- Since which gretl version has this been available? I'm asking because
of package requirements.
- How does the psdroot() function relate to all this?
thanks
sven
9:36 a.m.
On Fri, 26 Feb 2021, Sven Schreiber wrote:
Actually, I just checked section 17.6 of the guide, where it says
that
the matrix must be PSD if the power is fractional. (And must be
invertible --and thus PD-- if the power is negative.)
So maybe the whole issue is moot because it is already fixed and done?
Yes, as I now see, that's right.
Three follow-up questions on this:
- The power operation should be added to table 17.2 of the guide, not
only mentioned in the text of section 17.6.
Yes.
- Since which gretl version has this been available? I'm asking
because
of package requirements.
Since 2018d.
- How does the psdroot() function relate to all this?
The specialized psdroot is a good deal faster since it doesn't
depend on eigendecomposition.
Allin
9:37 a.m.
Am 26.02.2021 um 15:02 schrieb Sven Schreiber:
Am 26.02.2021 um 14:07 schrieb Allin Cottrell:
> On Fri, 26 Feb 2021, Riccardo (Jack) Lucchetti wrote:
>> On Fri, 26 Feb 2021, Sven Schreiber wrote:
>>> I just stumbled over old hansl code by Jack that implements the
>>> fractional power of a pd matrix (see at the end). The comment hints at
>>> the plan to integrate this somewhere, but it never happened.
OK, so
this "never happened" claim is refuted...
- Since which gretl version has this been available? I'm asking because
of package requirements.
Ah, the changelog is such a useful tool.... It says since
gretl 2018d.
- How does the psdroot() function relate to all this?
Dummy me; it produces triangular results, whereas the powers yield
symmetric results...
thanks
sven
8:34 a.m.
On Fri, 26 Feb 2021, Riccardo (Jack) Lucchetti wrote:
On Fri, 26 Feb 2021, Sven Schreiber wrote:
> Hi,
> I just stumbled over old hansl code by Jack that implements the
> fractional power of a pd matrix (see at the end). The comment hints at
> the plan to integrate this somewhere, but it never happened. Should it
> go into the 'extra' addon?
It's now in git, under the name "psdpwr" [...]
Where in git, please?
Allin
10:23 a.m.
On Fri, 26 Feb 2021, Allin Cottrell wrote:
On Fri, 26 Feb 2021, Riccardo (Jack) Lucchetti wrote:
> On Fri, 26 Feb 2021, Sven Schreiber wrote:
>
>> Hi,
>> I just stumbled over old hansl code by Jack that implements the
>> fractional power of a pd matrix (see at the end). The comment hints at
>> the plan to integrate this somewhere, but it never happened. Should it
>> go into the 'extra' addon?
>
> It's now in git, under the name "psdpwr" [...]
Where in git, please?
That's in the extra addon, but now that we've established that that
function is already implemewnted natively (dummy me, I should have
checked) I'm going to remove it.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
10:25 a.m.
On Fri, 26 Feb 2021, Riccardo (Jack) Lucchetti wrote:
> Where in git, please?
That's in the extra addon, but now that we've established that that function
is already implemewnted natively (dummy me, I should have checked) I'm going
to remove it.
Even better: I'd forgotten to "git push", so it was only in my local tree.
Sorry for the noise.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
1360
days inactive
1360
days old
8 comments
3 participants
participants (3)
-
Allin Cottrell -
Riccardo (Jack) Lucchetti -
Sven Schreiber