colorbraiding gives the wrong particles in error message

[jeanluct]

Margaux has a dataset that braidlab can't handle, and I can't see why. I attach the dataset.

[jeanluct]

From Jean-Luc Thiffeault on 2014-11-16 19:40:13+00:00

Note that this is a partial dataset containing only the first two time points.

[jeanluct]

From Jean-Luc Thiffeault on 2014-11-16 20:00:10+00:00

Ok, I understand the issue: since colorbraiding needs to sort the particles from left to right according to initial position, it returns an error message mentioning the wrong particles. In this case it claims particles 1 and 2 are coincident, when it's really 1 and 3.

I fixed this in colobraiding.m for the Matlab method. Marko: please fix in the C++ code. The way I did it requires passing the idx array. This is kind of ugly but doesn't really affect overhead. Maybe there's a better way? It's annoying but I think it's important that the error message refers to the right particles.

[jeanluct]

From Jean-Luc Thiffeault on 2014-11-16 20:00:55+00:00

Pass idx so that colorbraiding gives the right error msg. See issue #82.

→ <<cset 6fe24fe6fd40>>

[jeanluct]

Marko: was this fixed in the C++? The error message returning the wrong permutation?

[mbudisic]

Is there a dataset that I can use to test for this error? I cannot find it in my e-mails... Does it make sense to make a unit test for this?
I opened an iss82 branch for this issue.

[jeanluct]

I think it used to be attached to this issue but got lost in the migration from BitBucket. It's too bad that you can't attach things to a GitHub issue (except images).

Maybe I can just quote the values:

XY(:,:,1) =

   0.025000000000000  -1.025000000000000
   0.023723980025734  -1.024702440850582


XY(:,:,2) =

   0.025000000000000  -0.925000000000000
   0.025195202215646  -0.925081384217894


XY(:,:,3) =

   0.025000000000000  -1.025000000000000
   0.023723980025734  -1.024702440850582


XY(:,:,4) =

   0.025000000000000  -1.075000000000000
   0.023183108848995  -1.074611892145736

Without MEX, this generates the error

>> braid(XY)
Error using braidlab.braid/colorbraiding>crossingsToGenerators (line 186)
Particles 2 and 4 have coincident projection at time index 1: change projection
angle (type help braid.braid).

Error in braidlab.braid/colorbraiding (line 112)
    [gen,tcr,~] = crossingsToGenerators(XYtraj,t,idx);

Error in braidlab.braid (line 213)
        br = braidlab.braid.colorbraiding(b,1:size(b,1),secnd);

With MEX, we have

>> braid(XY)
Error using colorbraiding_helper
Particles 1 and 2 have coincident projection at time index 1: change projection
angle (type help braid.braid).
[Error 1/1 in threads.]

Error in braidlab.braid/colorbraiding (line 105)
  [gen,tcr] = colorbraiding_helper(XYtraj,t,Nthreads);

Error in braidlab.braid (line 213)
        br = braidlab.braid.colorbraiding(b,1:size(b,1),secnd);

So you see in the MEX version it gives the wrong particles.

If you change the projection angle, the error becomes coincident coordinates:

>> braid(XY,.1)
Error using colorbraiding_helper
Particles 2 and 3 have coincident coordinates at time index 1: braid not defined.
[Error 1/1 in threads.]

But it's 1 and 3 that have coincident coordinates.

[jeanluct]

Oh, and I don't think it's worth writing a unit test, since this is an error in the error message itself...

In fact it's esoteric enough that if it's going to take some work to fix, then we could kick it beyond 2.2. Your choice.

[mbudisic]

Here's the matrix with copy-paste supported.

XY = nan(2,2,4);
XY(:,:,1) = [0.025 -1.025;0.023723980025734 -1.02470244085058];
XY(:,:,2) = [0.025 -0.925;0.025195202215646 -0.925081384217894];
XY(:,:,3) = [0.025 -1.025;0.023723980025734 -1.02470244085058];
XY(:,:,4) = [0.025 -1.075;0.023183108848995 -1.07461189214574];

[mbudisic]

While I do understand what's going on (I think, at least), I get different errors:

global BRAIDLAB_braid_nomex; BRAIDLAB_braid_nomex = 1;
braidlab.braid(XY)

gives

Particles 4 and 1 have coincident projection at time index 1: change projection angle (type help braid.braid).

And it should be 1 and 3 right? Are different branches doing different things? Can you confirm that I understand things correctly?

global BRAIDLAB_braid_nomex; BRAIDLAB_braid_nomex = 0;
braidlab.braid(XY)

gives (incorrectly)

Particles 1 and 2 have coincident projection at time index 1: change projection angle (type help
braid.braid).

[jeanluct]

Maybe it has to do with re-reading the data: I loaded mine from a MEX file. Do you have access to the shared Dropbox folder with Margaux? It has a file called issue82.mat.

[mbudisic]

I don't, can you share it with me or e-mail me the file? It's small enough...

[jeanluct]

Not sure why I didn't just do that...

On Mon, Dec 8, 2014 at 3:32 PM, Marko Budisic notifications@github.com
wrote:

I don't, can you share it with me or e-mail me the file? It's small
enough...

—
Reply to this email directly or view it on GitHub
#82 (comment).

[mbudisic]

OK, I loaded the issue82.mat but it still looks like particles 1 and 3 are culprits, not 2 and 4 as Matlab code reports. Could the reason be that it's not the permutation vector that we want to pass, but rather its inverse?

[mbudisic]

Added devel/tests/test_issue82.m to demonstrate this.

[jeanluct]

No I think it works now: I added a projection angle of .1. The errors were mixed between coincident projection and coordinates, but if you rotate the projection angle there is no projection error, only coincident particles. The numbers seem to match now. (Committed on the same branch.)

[jeanluct]

I merged develop into iss82 and the testsuite doesn't work. Please check.

[mbudisic]

testsuite works on my end... (and it did before the commit too) I just pulled, fetched and everything.

Weren't you supposed to merge iss82 into develop (and not vice versa)?

[mbudisic]

Aha, without MEX it fails. I'll take a look.

[jeanluct]

I merge from develop to iss82 to see if it would solve the testsuite problem. It seems all I needed was a make clean. Works now.

Will merge into develop.