|
From: Lakshmi S. <lak...@ka...> - 2013-10-16 06:24:31
|
Hi Andy, I will see if I can fix it. Best regards Lakshmi. On Tue, Oct 15, 2013 at 3:49 PM, Andy Adler <ad...@sc...> wrote: > Hi Nolwenn, Hi Lakshmi, > > I would really like to see this function cleaned up, and made available as > a tutorial. It gives another way to approach absolute imaging, and may > be more stable in many cases. > > I just tried the UNIT_TEST, and the function isn't working (but it used > to). > Could you please try it, Nolwenn, and see if there is an obvious fix. > >> inv_solve_abs_CG_logc UNIT_TEST > > Thanks > -- > Andy Adler <ad...@sc...> +1-613-520-2600x8785 > > > On 15 October 2013 03:21, <les...@sc...> wrote: > > Dear Lakshmi, > > > > To choose the singular truncation value (later stored in the variable > > tol), I wrote the equation using the Hansen 1999 paper > > ( > http://www.sintef.no/upload/IKT/9011/SimOslo/vskoler/2005/notes/Lcurve.pdf > ). > > You'll find the equations 8 and 9 that describe XL2 and RES2. > > Indeed I compute the tol value only for the first iteration of my > > inversion and then I keep the same tol value. > > My tol value is not automatically determined with this function, since > you > > first need to explore the best interval of the Lambda values that will > > draw you an L that may not arise depending on your data quality, see: > > Mamatjan, Y., Borsic, A., Gürsoy, D., & Adler, A. (2013). An experimental > > clinical evaluation of EIT imaging with ℓ1 data and image norms. > > Physiological measurement, 34(9), 1027. > > > > What I usually do is: I run once the inversion with a large interval of > > lambda values (stored in imdl.parameters.lambda) and plot the L-curve. > > Then I run the inversion a second time with a lambda interval restricted > > to the region close to the L-curve I plot again the L-curve for choosing > > my tolerance value (on the corner of the L-curve or slightly on its right > > to increase the filter strength) and then I write it into : > > imdl.parameters.tol > > > > Since the problem may be non linear, I ran the inversion with something > > like 10 iterations, with different imdl.parameters.tol values around the > > one that seemed to be the best and I redraw an L-curve with the different > > norm of the residuals I found at the end of my inversions and the norm of > > the images roughness. I also compare the different resulting images, the > > presence of artifacts on the images... The choice of imdl.parameters.tol > > is of great importance, it depends on the geometry of your studied > object, > > the data quantity and quality... I don't think there is a simple way for > > determining its best value, the difficulty is the same for the > > determination of the hyperparameter values for the Gauss-Newton > > algorithms. > > > > Later in the code, the choice of imdl.parameters.perturb may require also > > some exploration... > > > > I hope that this could help you... > > > > Best regards, > > > > Nolwenn. > > > > > > > >> Hi Andy, > >> The version I am using is eidors v3.7.1. This is the file I found > > online > >> and you can see a function > >> function tol= svdAnalysisLcurvecrit(data,img,J) > >> < > http://eidors3d.sourceforge.net/doc/eidors/solvers/inverse/inv_solve_abs_CG_logc.html#_subfunctions > > > > in line number 189. My questions relate to the definitions within this > > function > >> particularly from the line 206-217. It is not under the deprecated > > folder and hence was wondering if this is not used at all for > >> the reconstructions. > >> Thanks. > >> Best regards > >> Lakshmi. > >> On Fri, Oct 11, 2013 at 2:36 PM, Andy Adler <ad...@sc...> > > wrote: > >>> Hi Lakshmi, > >>> We may be looking at a different file. The most recent version is in > > the > >>> web link I gave you, and doesn't have those lines. > >>> What version of EIDORS do you have? And what is the path to the file? > -- > >>> Andy Adler <ad...@sc...> +1-613-520-2600x8785 > >>> On 11 October 2013 02:39, Lakshmi Selvakumaran > >>> <lak...@ka...> wrote: > >>> > HI Andy, > >>> > The definitions are from the lines 206-217. My question refers to > > the > >>> > definition of XL2 and RES2. > >>> > Thanks. > >>> > Best regards > >>> > Lakshmi. > >>> > On Fri, Oct 11, 2013 at 3:41 AM, Andy Adler <ad...@sc...> > >>> wrote: > >>> >> Hi Lakshmi, > >>> >> The most recent version of the code is here: > >>> >> http://sourceforge. > >>> > .net/p/eidors3d/code/HEAD/tree/trunk/eidors/solvers/inverse/inv_solve_abs_GN_logC.m > >>> >> I'm having a little trouble seeing exactly where you have > >>> >> the question. Can you point out the line numbers and code? > >>> >> For example, we don't use variables with names beta or XL2. > >>> >> -- > >>> >> Andy Adler <ad...@sc...> +1-613-520-2600x8785 > >>> >> On 9 October 2013 10:07, Lakshmi Selvakumaran > >>> >> <lak...@ka...> wrote: > >>> >> > Hello, > >>> >> > Is there a reference that discusses the algorithm used in the > >>> function > >>> >> > 'inv_solve_abs_CG_logc'? > >>> >> > There are some things that I don't understand completely. If > >>> anyone > >>> can > >>> >> > explain, it will be of great help. > >>> >> > 1) In the SVD analysis, for the calculation of the XL2 variable, > > it > >>> is > >>> >> > not > >>> >> > multiplied by the right singular matrix, V. In literature, the x > >>> vector > >>> >> > is > >>> >> > the filter factor *( beta/singular values ) * vi. But in the > >>> function, > >>> >> > the > >>> >> > singular vectors vi is missing. Why? > >>> >> > 2) How is RES2 calculated? Again, it seems to me that some terms > >>> are > >>> >> > missing. > >>> >> > 3) Is the tol value that is calculated from the svd analysis > >>> represent > >>> >> > the > >>> >> > level of truncation in the singular values? > >>> >> > Thanks. > >>> >> > Best regards > >>> >> > Lakshmi. > >>> >> > ________________________________ > >>> >> > This message and its contents, including attachments are intended > >>> solely > >>> >> > for > >>> >> > the original recipient. If you are not the intended recipient or > >>> have > >>> >> > received this message in error, please notify me immediately and > >>> delete > >>> >> > this > >>> >> > message from your computer system. Any unauthorized use or > >>> distribution > >>> >> > is > >>> >> > prohibited. Please consider the environment before printing this > >>> email.. > >>> > ------------------------------------------------------------------------------ > >>> >> > October Webinars: Code for Performance > >>> >> > Free Intel webinars can help you accelerate application > >>> performance. > >>> >> > Explore tips for MPI, OpenMP, advanced profiling, and more. Get > > the > >>> most > >>> >> > from > >>> >> > the latest Intel processors and coprocessors. See abstracts and > >>> register > >>> > http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg..clktrk > >>> >> > _______________________________________________ > >>> >> > eidors3d-help mailing list > >>> >> > eid...@li... > >>> >> > https://lists.sourceforge.net/lists/listinfo/eidors3d-help > >>> > ------------------------------------------------------------------------------ > >>> >> October Webinars: Code for Performance > >>> >> Free Intel webinars can help you accelerate application performance. > > Explore tips for MPI, OpenMP, advanced profiling, and more. Get the > >>> most > >>> >> from > >>> >> the latest Intel processors and coprocessors. See abstracts and > >>> register > > >>> > http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg..clktrk > >>> >> _______________________________________________ > >>> >> eidors3d-help mailing list > >>> >> eid...@li... > >>> >> https://lists.sourceforge.net/lists/listinfo/eidors3d-help > >>> > ________________________________ > >>> > This message and its contents, including attachments are intended > >>> solely > >>> for > >>> > the original recipient. If you are not the intended recipient or have > > received this message in error, please notify me immediately and > >>> delete > >>> this > >>> > message from your computer system. Any unauthorized use or > >>> distribution > >>> is > >>> > prohibited. Please consider the environment before printing this > >>> email. > >>> > ------------------------------------------------------------------------------ > >>> > October Webinars: Code for Performance > >>> > Free Intel webinars can help you accelerate application performance. > > Explore tips for MPI, OpenMP, advanced profiling, and more. Get the > >>> most > >>> > from > >>> > the latest Intel processors and coprocessors. See abstracts and > >>> register > >>> > http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg..clktrk > >>> > _______________________________________________ > >>> > eidors3d-help mailing list > >>> > eid...@li... > >>> > https://lists.sourceforge.net/lists/listinfo/eidors3d-help > >>> > ------------------------------------------------------------------------------ > > October Webinars: Code for Performance > >>> Free Intel webinars can help you accelerate application performance. > > Explore tips for MPI, OpenMP, advanced profiling, and more. Get the > > most > >>> from > >>> the latest Intel processors and coprocessors. See abstracts and > > register > >>> > http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg..clktrk > > _______________________________________________ > >>> eidors3d-help mailing list > >>> eid...@li... > >>> https://lists.sourceforge.net/lists/listinfo/eidors3d-help > >> -- > >> ------------------------------ > >> This message and its contents, including attachments are intended solely > > for the original recipient. If you are not the intended recipient or > > have > >> received this message in error, please notify me immediately and delete > > this message from your computer system. Any unauthorized use or > >> distribution is prohibited. Please consider the environment before > printing > >> this email. > >> > ------------------------------------------------------------------------------ > > October Webinars: Code for Performance > >> Free Intel webinars can help you accelerate application performance. > > Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most > > from > >> the latest Intel processors and coprocessors. See abstracts and register > >> > >> > http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg.clktrk_______________________________________________ > > eidors3d-help mailing list > >> eid...@li... > >> https://lists.sourceforge.net/lists/listinfo/eidors3d-help > > > > > > > > > > > > > ------------------------------------------------------------------------------ > > October Webinars: Code for Performance > > Free Intel webinars can help you accelerate application performance. > > Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most > from > > the latest Intel processors and coprocessors. See abstracts and register > > > > > http://pubads.g.doubleclick.net/gampad/clk?id=60135031&iu=/4140/ostg.clktrk > > _______________________________________________ > > eidors3d-help mailing list > > eid...@li... > > https://lists.sourceforge.net/lists/listinfo/eidors3d-help > -- ------------------------------ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email. |