<html>
<head>
<meta content="text/html; charset=utf-8" http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">Le 19/05/2015 22:50, Jim Bosch a
écrit :<br>
</div>
<blockquote
cite="mid:CAN+0q25X3xcqYAeeiTHE2W+OfTo5wXCOXE3ewg9-s2VfH8WS0g@mail.gmail.com"
type="cite">
<div dir="ltr">I think this looks like a great start, and I do
think it'd be a good idea to start integrating it into the stack
as soon as possible (though might depend on how much DM effort
is available to help with that). I mostly agree with your ideas
for other "next steps" - though I don't think proper
motion/parallax needs to be a top priority, and I do wonder how
much work it'd be to get better outlier rejection (would we have
to write our own custom low-rank Cholesky update?)</div>
</blockquote>
Hi Jim,<br>
<br>
Regarding small rank updates, I still have to figure out if this
cholmod capability is provided<br>
through the Eigen interface to Cholmod. If it is the case, It seems
to me that a rank-2 update can<br>
be turned into 2 rank-1 updates and this could be the easy way. I
have not tried seriously<br>
to talk directly to cholmod in C(++). One way I am considering is to
switch to python <br>
once we have the derivatives, and call all sparse algebra utilities
from python. Marc and Nicolas<br>
tell me that it is easy. One extra incentive to do so is that for
non linear fits, the line search<br>
is much easier to set up in python than in C++, and it will not make
any difference for the execution<br>
time, as long as traversing the measurements runs in C++. <br>
One thing I have not considered so far is to implement this
capability into some existing <br>
factorization code. It might turn out to be difficult. <br>
<br>
However, if the aim is to implement this simultaneous
astrometry into the stack <br>
ASAP (or sort of), efficiently clipping outliers is not a "must
have".<br>
<br>
<blockquote
cite="mid:CAN+0q25X3xcqYAeeiTHE2W+OfTo5wXCOXE3ewg9-s2VfH8WS0g@mail.gmail.com"
type="cite">
<div dir="ltr">
<div><br>
</div>
<div>I imagine there will eventually be a lot of modification to
what your classes look like now in order to integrate them
with existing LSST classes with overlapping or related
functionality. But it might make the most sense to just
ignore most of that for now and just move this into a new LSST
package immediately, write some adapters to make it accept
inputs from the rest of the LSST pipeline, and then integrate
those individual classes one at a time later.</div>
<div><br>
</div>
</div>
</blockquote>
<br>
If time is a concern, this is the way to go. That's also much easier
for me! Then, my next step<br>
should be to dive again into the stack code and figure out mostly
what are the interfaces to <br>
the catalogs and Fits headers we should be using for that. I have
not figured out yet what<br>
the forces are in France for this effort. We are having a meeting
next week where the question <br>
will be raised. <br>
<br>
<br>
Pierre.<br>
<br>
<br>
<br>
<blockquote
cite="mid:CAN+0q25X3xcqYAeeiTHE2W+OfTo5wXCOXE3ewg9-s2VfH8WS0g@mail.gmail.com"
type="cite">
<div dir="ltr">
<div><br>
</div>
<div>Jim</div>
<div><br>
</div>
<div><br>
</div>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">On Tue, May 19, 2015 at 11:32 AM,
Pierre Astier <span dir="ltr"><<a moz-do-not-send="true"
href="mailto:pierre.astier@in2p3.fr" target="_blank">pierre.astier@in2p3.fr</a>></span>
wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">Hi,<br>
<br>
Since our first phone conference about the development of
a simultaneous<br>
astrometry package, I have coded a fit and ran a couple of
tests on exposure<br>
sets from the CFHTLS and Suprime cam. I have described the
work and the first<br>
results in a few slides that you can find at:<br>
<a moz-do-not-send="true"
href="http://supernovae.in2p3.fr/%7Eastier/sim-ast.pdf"
target="_blank">http://supernovae.in2p3.fr/~astier/sim-ast.pdf</a><br>
<br>
I am mostly wondering whether this is enough to start
discussing<br>
how we practically implement the algorithms into the stack.<span
class="HOEnZb"><font color="#888888"><br>
<br>
Pierre.<br>
<br>
-- <br>
-------------------<br>
Pierre Astier , LPNHE, 12-22 1er étage.<br>
4 place Jussieu, F 75252 Paris Cedex 05<br>
tel (33) 1 44 27 76 47 ---- fax (33) 1 44 27 46 38<br>
<br>
_______________________________________________<br>
dm-astrometry mailing list<br>
<a moz-do-not-send="true"
href="mailto:dm-astrometry@lists.lsst.org"
target="_blank">dm-astrometry@lists.lsst.org</a><br>
<a moz-do-not-send="true"
href="https://lists.lsst.org/mailman/listinfo/dm-astrometry"
target="_blank">https://lists.lsst.org/mailman/listinfo/dm-astrometry</a><br>
</font></span></blockquote>
</div>
<br>
</div>
</blockquote>
<br>
<br>
<pre class="moz-signature" cols="72">--
-------------------
Pierre Astier , LPNHE, 12-22 1er étage.
4 place Jussieu, F 75252 Paris Cedex 05
tel (33) 1 44 27 76 47 ---- fax (33) 1 44 27 46 38
</pre>
</body>
</html>