Re: [GSoC 2019] Matrix pseudoinverse and Least squares solver

David Bellot via ublas <[email protected]> Fri, 12 Apr 2019 14:10:47 +1000
Newsgroups gmane.comp.lib.boost.ublas
Message-ID <CAOE6ZJFOVBJN+3pizL1oviRfkcEbUJbw8n9BRL8MweHvY3DPNA@mail.gmail.com>
--===============1519796692827048151==
Content-Type: multipart/alternative; boundary="00000000000058329c05864d7e78"

--00000000000058329c05864d7e78
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hi,

we have a solver algorithms done in GSOC which is not yet integrated into
the main branch. I think we should start from there. The GSOC was very good
and the algorithm stable. So if we have it integrated into the main branch,
then implementing least-square solvers and other algorithm will be very
easy.

I'll let you have a look at the algo here:
https://github.com/BoostGSoC15/ublas/tree/master/include/boost/numeric/ubla=
s

On Tue, Apr 2, 2019 at 4:10 AM Thomas Yang via ublas <[email protected]=
>
wrote:

> Hi all,
>
> I was investigating some methods to perform least squares algorithms, and
> I noticed we currently do not have methods of doing so in the development
> branch.
>
> To the extent of my knowledge, we also do not have a method of determinin=
g
> matrix inverses or pseudoinverse. All of the existing implementations I
> found (e.g. rTensor) of CP decomposition have a method of determining a
> matrix's pseudoinverse.
>
> For GSoC, I initially based my proposal around implementing tensor
> decomposition algorithms. However, would it be a better choice to focus
> more on least squares methods like QR-factorization=E2=80=9D And SVD? Fur=
thermore,
> was this already attempted in a past GSoC project?
>
> Thanks,
>
> Thomas
> _______________________________________________
> ublas mailing list
> [email protected]
> https://lists.boost.org/mailman/listinfo.cgi/ublas
> Sent to: [email protected]
>

--00000000000058329c05864d7e78
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr"><div class=3D"gmail_default" style=3D"fon=
t-family:arial,helvetica,sans-serif;font-size:small">Hi,</div><div class=3D=
"gmail_default" style=3D"font-family:arial,helvetica,sans-serif;font-size:s=
mall"><br></div><div class=3D"gmail_default" style=3D"font-family:arial,hel=
vetica,sans-serif;font-size:small">we have a solver algorithms done in GSOC=
 which is not yet integrated into the main branch. I think we should start =
from there. The GSOC was very good and the algorithm stable. So if we have =
it integrated into the main branch, then implementing least-square solvers =
and other algorithm will be very easy.</div><div class=3D"gmail_default" st=
yle=3D"font-family:arial,helvetica,sans-serif;font-size:small"><br></div><d=
iv class=3D"gmail_default" style=3D"font-family:arial,helvetica,sans-serif;=
font-size:small">I&#39;ll let you have a look at the algo here:=C2=A0<a hre=
f=3D"https://github.com/BoostGSoC15/ublas/tree/master/include/boost/numeric=
/ublas">https://github.com/BoostGSoC15/ublas/tree/master/include/boost/nume=
ric/ublas</a></div></div></div><br><div class=3D"gmail_quote"><div dir=3D"l=
tr" class=3D"gmail_attr">On Tue, Apr 2, 2019 at 4:10 AM Thomas Yang via ubl=
as &lt;<a href=3D"mailto:[email protected]">[email protected]</a>&g=
t; wrote:<br></div><blockquote class=3D"gmail_quote" style=3D"margin:0px 0p=
x 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div d=
ir=3D"ltr">Hi all,<br><br>I was investigating some methods to perform least=
 squares algorithms, and I noticed we currently do not have methods of doin=
g so in the development branch.=C2=A0<br><br>To the extent of my knowledge,=
 we also do not have a method of determining matrix inverses or pseudoinver=
se. All of the existing implementations I found (e.g. rTensor) of CP decomp=
osition have a method of determining a matrix&#39;s pseudoinverse.=C2=A0<br=
><br>For GSoC, I initially based my proposal around implementing tensor dec=
omposition algorithms. However, would it be a better choice to focus more o=
n least squares methods like QR-factorization=E2=80=9D And SVD? Furthermore=
, was this already attempted in a past GSoC project?<br><br>Thanks,<br><br>=
Thomas=C2=A0=C2=A0<br></div>
_______________________________________________<br>
ublas mailing list<br>
<a href=3D"mailto:[email protected]" target=3D"_blank">[email protected]=
t.org</a><br>
<a href=3D"https://lists.boost.org/mailman/listinfo.cgi/ublas" rel=3D"noref=
errer" target=3D"_blank">https://lists.boost.org/mailman/listinfo.cgi/ublas=
</a><br>
Sent to: <a href=3D"mailto:[email protected]" target=3D"_blank">david.=
[email protected]</a><br>
</blockquote></div>

--00000000000058329c05864d7e78--

--===============1519796692827048151==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline