 vasiloglou
|
ef2fdfb8c3
|
the bug in the gradient was found now it is working but it seems to me that the minimum has a lot of points
|
2008-05-27 06:14:27 +00:00 |
|
|
Dongryeol Lee
|
ce471f9889
|
Current stable version.
|
2008-05-27 00:54:13 +00:00 |
|
|
vasiloglou
|
322e57a959
|
some problems resolved, the barrier is not working properly, there is the problem
|
2008-05-26 08:28:02 +00:00 |
|
|
vasiloglou
|
f660925df4
|
the convex mvu is sort of working not sure yet, the results are poor
|
2008-05-26 05:39:34 +00:00 |
|
|
vasiloglou
|
89bdb6f565
|
extended copy for matrices
|
2008-05-26 05:38:38 +00:00 |
|
|
Dongryeol Lee
|
e5227ed824
|
Fixed the bug that caused FurthestColumnIndex column to return -1 as the index.
|
2008-05-26 03:03:39 +00:00 |
|
|
vasiloglou
|
70b4c58484
|
it is running for a small matrix with 31% error and it seems to me something is
wrong, LBFGS keeps failing, while it shouldn't
|
2008-05-26 02:38:15 +00:00 |
|
|
Dongryeol Lee
|
f058da938d
|
Minor bug fix plus now the far-field expansion of the matrix-factorized version stores the maximum incurred error in approximation.
|
2008-05-25 19:56:53 +00:00 |
|
|
Dongryeol Lee
|
e7c208d13f
|
Works reasonable well - now need to enforce that the local moments are actually pushed downwards whenever the query tree is recursed so that we get a tighter lower bound.
|
2008-05-25 19:56:00 +00:00 |
|
|
vasiloglou
|
b5ba2d5d57
|
now it looks more stable but doesn't converge yet
i will better check it with a smaller matrix
Probably the gradient is wrong
|
2008-05-25 04:44:18 +00:00 |
|
|
vasiloglou
|
44ebad5424
|
gradient gets nan, probably because somethins is unitialized
|
2008-05-24 22:20:04 +00:00 |
|
|
vasiloglou
|
c42b64efb8
|
compiled
|
2008-05-24 21:07:42 +00:00 |
|
|
Parikshit Ram
|
68c74490b2
|
updating main and optimizers_reloaded.h
|
2008-05-24 21:03:50 +00:00 |
|
|
Parikshit Ram
|
0a1cc530e8
|
SMD multi step added with fast Hv product in main
|
2008-05-24 20:59:42 +00:00 |
|
|
Parikshit Ram
|
b779c59708
|
distances changed and exptl folder added
|
2008-05-24 20:54:05 +00:00 |
|
|
vasiloglou
|
167af932f7
|
addded more files
|
2008-05-24 19:58:44 +00:00 |
|
|
vasiloglou
|
e510a16978
|
Our last attempt to make sdp nmf work
|
2008-05-24 03:22:56 +00:00 |
|
|
jwaters6
|
c46a49b7dd
|
Changed kdtree_impl functions to pass by reference (when applicable)
|
2008-05-23 19:57:54 +00:00 |
|
|
jwaters6
|
a2e5f9041c
|
Revised kdtree & bounds updates to merge changes
|
2008-05-23 16:06:54 +00:00 |
|
|
rriegel
|
e0e859b901
|
todo at top of allnn.h
|
2008-05-23 08:21:33 +00:00 |
|
|
vasiloglou
|
985f0b68ed
|
changes name
|
2008-05-23 04:48:36 +00:00 |
|
|
vasiloglou
|
b5b15f16c3
|
It turns out SDP with rank one gives the same error with just simple NMF
|
2008-05-23 04:42:04 +00:00 |
|
|
Dongryeol Lee
|
a10d6655d6
|
A minor change to the code that involves how to compute the ending index of the points owned by the node.
|
2008-05-23 00:54:49 +00:00 |
|
|
Dongryeol Lee
|
031150ba59
|
A little tweak in the number of samples to take in CUR decomposition - this needs to be automated!
|
2008-05-23 00:47:35 +00:00 |
|
|
Dongryeol Lee
|
6dffee972a
|
Bug fix in forming the outgoing representation - incorrect matrix-vector multiplication has been fixed.
|
2008-05-23 00:12:23 +00:00 |
|
|
Dongryeol Lee
|
0a75b435e5
|
Reverted back to the old CUR version... Now need to debug...
|
2008-05-22 23:35:16 +00:00 |
|
|
vasiloglou
|
4976c5e848
|
Put the code in order, and ready to play with the original formulation
|
2008-05-22 23:05:00 +00:00 |
|
|
Dongryeol Lee
|
0cd9faa025
|
A segmentation fault fix.
|
2008-05-22 13:48:10 +00:00 |
|
|
vasiloglou
|
d9ae0ec7ca
|
it seems to run very well for rank one only although I haven't checked the results
|
2008-05-22 04:43:58 +00:00 |
|
|
Dongryeol Lee
|
90e4d29e3b
|
A minor fix to the matrix-factorized FMM. Still need to figure out how to do error bounds.
|
2008-05-22 02:40:53 +00:00 |
|
|
Dongryeol Lee
|
5b2c49e878
|
Updated the matrix-factorized fmm to use exact cur decomposition with pivotting qr.
|
2008-05-22 02:39:35 +00:00 |
|
|
Bill March
|
4d6942ba98
|
fixed symmetry and hybrid error
|
2008-05-21 18:27:00 +00:00 |
|
|
vasiloglou
|
c48abf6bad
|
it is running now!!!!!
|
2008-05-21 14:28:15 +00:00 |
|
|
vasiloglou
|
0079c3a637
|
it seems to work , needs to add an svd on the results, svd on matlab showed positive eigen_vector
|
2008-05-21 04:58:07 +00:00 |
|
|
vasiloglou
|
5fec2cbf11
|
fixed the error in the gradient now minimizes
|
2008-05-21 04:39:49 +00:00 |
|
|
Dongryeol Lee
|
afe908b6de
|
Some minor fixes to the matrix factorized series expansion.
|
2008-05-21 01:32:12 +00:00 |
|
|
Dongryeol Lee
|
df18fbdad0
|
It works sort of - I need to figure out how to do automatic error control at least probabilistically.
|
2008-05-21 01:31:46 +00:00 |
|
|
vasiloglou
|
50acdc0af0
|
compiled but doesn't run properly yet
|
2008-05-20 22:43:01 +00:00 |
|
|
jwaters6
|
40867b44cc
|
|
2008-05-20 19:44:34 +00:00 |
|
|
jwaters6
|
cd55332024
|
|
2008-05-20 19:38:58 +00:00 |
|
|
vasiloglou
|
388453d943
|
close to compiling
|
2008-05-20 19:14:52 +00:00 |
|
|
vasiloglou
|
6b59a6daff
|
added more files
|
2008-05-20 18:20:28 +00:00 |
|
|
vasiloglou
|
4d65a5c7e7
|
first commit
|
2008-05-20 03:17:13 +00:00 |
|
|
Dongryeol Lee
|
f0d1921623
|
Need to implement the canonical dual-tree recursion and debug...
|
2008-05-20 02:18:35 +00:00 |
|
|
tekhnofiend
|
561c985bad
|
|
2008-05-20 02:18:26 +00:00 |
|
|
Dongryeol Lee
|
b5ec42b29b
|
Implemented the evaluation method for matrix-factorized local expansion.
|
2008-05-20 02:18:08 +00:00 |
|
|
Dongryeol Lee
|
bec053e9b4
|
Forgot to initialize the local moments - now wrote the code for doing so.
|
2008-05-20 01:08:49 +00:00 |
|
|
Dongryeol Lee
|
6015ec3a94
|
More compilation error fix.
|
2008-05-20 00:51:36 +00:00 |
|
|
Dongryeol Lee
|
1d2c20a944
|
More compilation error fix.
|
2008-05-20 00:51:10 +00:00 |
|
|
Dongryeol Lee
|
368328b629
|
Added the wrapper kernel class for the matrix-factorized FMM.
|
2008-05-19 20:21:18 +00:00 |
|