Prev: Use of kstest() function in Matlab
Next: Sort vector with out using sort function
From: Eric Diaz on 26 Apr 2010 11:43
Dear Bruno,

Clearly, your thoughts are biased as well. There is no turning about a bump...that is preposterous. The purpose of a bump is well known by all; It is to refresh a posts page status (i.e. back to page 1) on a forum which is very active. There was a great deal of religious and merchandise spam on the forum, and thus the reason for bumping. It was unfortunate that someone was tasteless enough to call my post spam for bumping.

I further disagree about John leaving the door open...if by being vague you mean... then of course he begs the question or interpretation, rather than actually being helpful. The only logical question that should have followed is "where exactly in your optim tips: did you discuss how to create confidence intervals for this specific problem?" Which of course, the answer to would be nowhere because this specific problem has not been addressed.

I believe my original post is clear and my questions are clear. I further posted up an specific example setup where I stated exactly what I would like to change with respect to that setup using variable projection. If you don't get it then I'm not sure what to do.

As to your question, I like to use nlpredci's delta to plot finely spaced confidence intervals around my fit curve. I tried looking at the nlpredci code but it was a bit jumbled. Otherwise, if I could compute the delta another way, I would be perfectly satisfied with my sensitivity analysis.

Eric Diaz
From: Bruno Luong on 26 Apr 2010 12:58
"Eric Diaz" <eric.diaz(a)gmail.com> wrote in message <hr4ca8$m9d$1(a)fred.mathworks.com>...
> Clearly, your thoughts are biased as well. There is no turning about a bump...that is preposterous. The purpose of a bump is well known by all; It is to refresh a posts page status (i.e. back to page 1) on a forum which is very active.
>There was a great deal of religious and merchandise spam on the forum,
> and thus the reason for bumping.

Correct me if I'm wrong but Eric:
- A "bump" is mainly used to tell "I'm not happy my request is ignored", isn't it?
- Can you imagine a mess created if every real thread would get a "bump refresh" everytime it get superposed by a spam post?

You are the first one who used "bump" in *that* purpose, that might be missinterpreted by many of us, myself included.

But honestly I don't think I missinterpreted: you obviously get frustrated that John's answer did not meet your expectation - as you have expresssed later, not only because of the string of spams.

> It was unfortunate that someone was tasteless enough to call my post spam for bumping.

The unfortunate tagging might be just a mistake, not intentionaly. It happens to me to tag inadvertant spam a true post.

>
> I further disagree about John leaving the door open...if by being vague you mean... then of course he begs the question or interpretation, rather than actually being helpful. The only logical question that should have followed is "where exactly in your optim tips: did you discuss how to create confidence intervals for this specific problem?" Which of course, the answer to would be nowhere because this specific problem has not been addressed.
>
> I believe my original post is clear and my questions are clear. I further posted up an specific example setup where I stated exactly what I would like to change with respect to that setup using variable projection. If you don't get it then I'm not sure what to do.

Personaly I did not (and still do no) get the GOAL you want to reach with the description. It seems to me you want to feed the full Jacobian somewhere in a specific method of solving least-square linear separable. But you have not explained clearly "why".

>
> As to your question, I like to use nlpredci's delta to plot finely spaced confidence intervals around my fit curve. I tried looking at the nlpredci code but it was a bit jumbled. Otherwise, if I could compute the delta another way, I would be perfectly satisfied with my sensitivity analysis.

I do not have the stat toolbox, so I can't comment.

Bruno
From: Walter Roberson on 26 Apr 2010 13:25
Eric Diaz wrote:
> Dear Walter,

> This forum does not belong to you, nor does it make you
> important, nor can you go around harassing people about their posts.

The wording of Eric Diaz's sentence, although semantically null in its
detailed phrasing, could have been reasonably _interpreted_ by some
people as being an indication that I have been "harassing" people, or
that Eric Diaz believed or believes I have been "harassing" people.

"Harassment" is a crime in Canada, a violation of Canada Criminal Code
section 264,
http://laws.justice.gc.ca/eng/C-46/20100425/page-6.html#codese:264 .

I do not believe that I have "harassed" anyone at any time within the
meaning of Canadian law -- a law that as a principle component requires
behaviour "that causes that other person reasonably, in all the
circumstances, to fear for their safety or the safety of anyone known to
them."

I have, though, engaged in what I believe to be "Fair comment on public
person or work of art", section 310(a),
http://laws.justice.gc.ca/eng/C-46/20100425/page-6.html#codese:310 .


Note: this response is not issued in contravention of section 264(2)(b),
but rather is published under the authority of Canada Criminal Code
section 312(b) "Publication invited or necessary"
http://laws.justice.gc.ca/eng/C-46/20100425/page-6.html#codese:312


I would suggest that correspondence (if any) on this matter proceed
through email, rather than in public.

Walter Roberson
From: Eric Diaz on 26 Apr 2010 14:48
Dear Bruno,

> Correct me if I'm wrong but Eric:
> - A "bump" is mainly used to tell "I'm not happy my request is ignored", isn't it?
> - Can you imagine a mess created if every real thread would get a "bump refresh" everytime it get superposed by a spam post?

You are wrong Bruno. Consider that my correction. It was used in the purpose that I previously stated and is more commonly used in that sense than as a way to whine about posts being ignored.

I was frustrated with the lack of usefullness of John's comment but I endeavored to reach a more concrete explanation by providing an example. Unfortunately, the entirety of this post's main concern was sidetracked by some cheeky comments.

I believe you misunderstood the spam comment. It wasn't that somebody accidentally mistagged my post as spam, it was that walter called my post spam in his reply. Which I took offense to.

> Personaly I did not (and still do no) get the GOAL you want to reach with the description. It seems to me you want to feed the full Jacobian somewhere in a specific method of solving least-square linear separable. But you have not explained clearly "why".
>
Why I would like to feed in the jacobian into my least squares nonlinear fit, should be obvious, e.g. fewer iterations. The variable projection strategy in the exponential case reduces the computational cost of a nonlinear fit of multiple parameters and perhaps also the number of iterations, while providing a similar if not smaller residual norm. Making it usually a more efficient and robust technique.

Eric Diaz
From: Bruno Luong on 26 Apr 2010 16:25
"Eric Diaz" <eric.diaz(a)gmail.com> wrote in message <hr4n56$l34$1(a)fred.mathworks.com>...

> Why I would like to feed in the jacobian into my least squares nonlinear fit, should be obvious, e.g. fewer iterations. The variable projection strategy in the exponential case reduces the computational cost of a nonlinear fit of multiple parameters and perhaps also the number of iterations, while providing a similar if not smaller residual norm. Making it usually a more efficient and robust technique.
>

It is still not entirely clear to me what is the issue.

1- Are you asking how to compute the Jacobian of the non-linear projection? It is obvious that the Jacobian where users must provide is the derivative of the model with respect to the optimized parameter(s), and nothing more.
2- If yes, do you know how to compute the Jacobian in 1 above?
3- If no, do you know how to compute the Jacobian of the original problem (i.e., without projection, I guess the answer is yes that this point from what I read)?
4- If yes are you able to compute 2 from 3?
5- If no have you heard about Golub/Pereyra formula?
6- If no try to get this paper: "G. H. Golub and V. Pereyra, The differentiation of pseudoinverses and nonlinear least squares problems whose variables separate, SIAM J. Numer. Anal., 10 (1973), pp. 413&#8211;432.'

Bruno
First  |  Prev  |  Next  |  Last
Pages: 1 2 3
Prev: Use of kstest() function in Matlab
Next: Sort vector with out using sort function