Skip to main content
22 events
when toggle format what by license comment
S Aug 27, 2019 at 14:00 history suggested Misi CC BY-SA 4.0
Link corrected
Aug 27, 2019 at 7:32 review Suggested edits
S Aug 27, 2019 at 14:00
Jun 12, 2014 at 7:24 history bounty awarded Stephan Kolassa
Jun 7, 2014 at 16:26 history edited Eamon Nerbonne CC BY-SA 3.0
added 234 characters in body
Jun 7, 2014 at 16:24 comment added Eamon Nerbonne @StephanKolassa: I've also included a sample implementation now. Note that your example input is actually a nasty case because it violates the triangle inequality...
Jun 7, 2014 at 16:09 comment added Stephan Kolassa @EamonNerbonne: sorry, my bad. I had misunderstood the definition of J. I worked through the PDF you linked, and everything is clear now. May I suggest that you edit your answer to include the link? And once again: thank you for your help!
Jun 7, 2014 at 15:48 history edited Eamon Nerbonne CC BY-SA 3.0
Added a simple implementation.
Jun 7, 2014 at 15:21 comment added Eamon Nerbonne JDJ will not have a zero diagonal when D has a zero diagonal - what's the reasoning there? I just implemented this to verify, and it works. I'll include the code in the answer.
Jun 7, 2014 at 10:05 vote accept Stephan Kolassa
Jun 7, 2014 at 10:05 comment added Stephan Kolassa There still seems to be something off. B=X^tX will have the squared lengths of the original vectors on the diagonal, but DD, as a distance matrix, will have a zero diagonal, so JDJ will also have a zero diagonal, so B can't be equal to JDJ. But you have definitely answered my question, given the Wikipedia entry and the pdf you linked to in your comment, so I'll go ahead and accept. Thanks a lot!
Jun 6, 2014 at 15:57 comment added Eamon Nerbonne @StephanKolassa: that better?
Jun 6, 2014 at 15:52 history edited Eamon Nerbonne CC BY-SA 3.0
added 8 characters in body
Jun 6, 2014 at 15:25 comment added Eamon Nerbonne @StephanKolassa: pages 9-11 here give a good example: homepages.uni-tuebingen.de/florian.wickelmaier/pubs/…. I'll update my answer shortly with the missing step once I figure out how to intuitively explain it :-)
Jun 6, 2014 at 15:14 comment added Eamon Nerbonne @StephanKolassa Oh right, I forgot the centering step, let me look that up again...
Jun 6, 2014 at 14:59 comment added Stephan Kolassa This does look promising. However, I kind of get stuck at the distance matrix being the product of the transpose of the position matrix with itself. I can't get that to work for simple examples. Given that the crossproduct is not translation-invariant, while the distance certainly is, I am sure I am missing something. Could you perhaps elaborate a bit?
Jun 6, 2014 at 14:20 comment added Eamon Nerbonne @michipili It happens to be exactly the problem I've encountered before, and a really nifty trick that turns out to be applicable more often than you might think :-)
Jun 6, 2014 at 14:18 history edited Eamon Nerbonne CC BY-SA 3.0
add example
Jun 6, 2014 at 14:16 comment added Michaël Le Barbier A very fair answer indeed!
Jun 6, 2014 at 14:04 review First posts
Jun 6, 2014 at 14:37
Jun 6, 2014 at 14:03 history edited Eamon Nerbonne CC BY-SA 3.0
added 659 characters in body
Jun 6, 2014 at 13:55 history edited Eamon Nerbonne CC BY-SA 3.0
added 659 characters in body
Jun 6, 2014 at 13:46 history answered Eamon Nerbonne CC BY-SA 3.0