 Dongryeol Lee
|
3ec5a9567d
|
Got it to compile
|
2008-02-22 01:12:55 +00:00 |
|
|
Dongryeol Lee
|
ab629b862c
|
More fix to come for krylov_lpr_solver_impl.h
|
2008-02-21 21:54:32 +00:00 |
|
|
Dongryeol Lee
|
6e58cc1499
|
Separated out heuristic function for node expansion into a separate file
|
2008-02-21 20:52:48 +00:00 |
|
|
Dongryeol Lee
|
534c8b5c14
|
Added a utility function for LPR
|
2008-02-21 20:35:13 +00:00 |
|
|
Dongryeol Lee
|
0f24fcf7be
|
Started debugging the krylov solver part
|
2008-02-21 20:34:33 +00:00 |
|
|
Dongryeol Lee
|
0d0ca195b6
|
Went through up to the phase 1 part of the matrix free algorithm
|
2008-02-21 20:23:02 +00:00 |
|
|
vasiloglou
|
b616334701
|
Added a maximum cap for the infeasibility
Added another mode where we maximize the furthest neighbors
|
2008-02-21 19:53:56 +00:00 |
|
|
Dongryeol Lee
|
57e2ccf965
|
Fix in the bug in pruning
|
2008-02-21 19:46:51 +00:00 |
|
|
Dongryeol Lee
|
2026282ca9
|
A minor fix
|
2008-02-21 19:22:41 +00:00 |
|
|
Dongryeol Lee
|
7deebec6cb
|
Minor bug fix
|
2008-02-21 18:47:22 +00:00 |
|
|
vasiloglou
|
872e7b492d
|
|
2008-02-21 06:02:21 +00:00 |
|
|
Dongryeol Lee
|
c51e54c8a8
|
Almost done reviving the matrix-free version of LPR
|
2008-02-21 03:37:30 +00:00 |
|
|
Dongryeol Lee
|
19cbfb88df
|
Second round of going through and fixing compilation errors
|
2008-02-21 03:08:34 +00:00 |
|
|
Dongryeol Lee
|
a8d8fae293
|
Renamed and restructured the matrix-free version
|
2008-02-21 02:18:54 +00:00 |
|
|
vasiloglou
|
6cf67f1082
|
changed from armijo to wolfe and now it works it is much more stable
We are ready to go to 10000 now
|
2008-02-20 10:46:46 +00:00 |
|
|
Dongryeol Lee
|
296aeff1c4
|
Took out EpanKernel moment out and made it a separate header
|
2008-02-20 02:50:02 +00:00 |
|
|
Dongryeol Lee
|
09a5d16c09
|
Some compiling error fix for the matrix-free driver
|
2008-02-20 02:28:52 +00:00 |
|
|
Dongryeol Lee
|
ac0831b7dd
|
Naive algorithm now uses much less memory
|
2008-02-20 01:19:15 +00:00 |
|
|
Dongryeol Lee
|
3eac4327e9
|
Single tree mode added
|
2008-02-19 21:36:42 +00:00 |
|
|
Dongryeol Lee
|
a7b3597053
|
Methods added to the main driver
|
2008-02-19 19:42:11 +00:00 |
|
|
Dongryeol Lee
|
3abe70ef77
|
Added variable bandwidth support
|
2008-02-19 19:33:26 +00:00 |
|
|
vasiloglou
|
09fd227005
|
|
2008-02-19 17:57:16 +00:00 |
|
|
vasiloglou
|
c26f42def1
|
this one seems to work
Unfortunatelly if we start randomly then we get nonsense
|
2008-02-19 07:58:16 +00:00 |
|
|
Dongryeol Lee
|
d3de9a182f
|
Final performance tuning on the dense version of the algorithm
|
2008-02-19 05:27:35 +00:00 |
|
|
Dongryeol Lee
|
8ac34d0683
|
Now does proper pruning based on the B^T W(q)^2 B component, with Epanechnikov kernel acceleration
|
2008-02-19 04:32:39 +00:00 |
|
|
Dongryeol Lee
|
dde0dd0233
|
Faster Epanechnikov pruning added
|
2008-02-19 03:00:53 +00:00 |
|
|
Dongryeol Lee
|
ebadacf9be
|
Change in bound computing function
|
2008-02-19 02:53:02 +00:00 |
|
|
Dongryeol Lee
|
994a960cf9
|
Dense LPR cannot run beyond linear order
|
2008-02-18 21:53:29 +00:00 |
|
|
Dongryeol Lee
|
e4adc8a271
|
Added the lower bound optimization for the dense formulation
|
2008-02-18 21:29:29 +00:00 |
|
|
Dongryeol Lee
|
4de9e7109a
|
Some performance tuning
|
2008-02-18 19:05:46 +00:00 |
|
|
Dongryeol Lee
|
65bf7b1c36
|
Minor variable name fix
|
2008-02-18 02:32:10 +00:00 |
|
|
Dongryeol Lee
|
99c6e39d80
|
Epanechnikov series expansion added for code
|
2008-02-18 02:10:06 +00:00 |
|
|
vasiloglou
|
6d6f66f78b
|
It works suspiciously
I am not sure it is right yet
|
2008-02-17 23:33:10 +00:00 |
|
|
Dongryeol Lee
|
bbee8bebaf
|
Fast code now computes confidence bands
|
2008-02-17 04:36:38 +00:00 |
|
|
Dongryeol Lee
|
2e9e51b9c8
|
No bug - just a misuse of the kernel function in two algorithms during comparison
|
2008-02-17 02:47:40 +00:00 |
|
|
Dongryeol Lee
|
56a601f145
|
Some bug fix
|
2008-02-17 02:39:00 +00:00 |
|
|
vasiloglou
|
3c3510610c
|
it doesn't give the expected results on the swiss roll
|
2008-02-17 00:16:49 +00:00 |
|
|
vasiloglou
|
3dbdd1d114
|
it is terminating now
|
2008-02-16 06:33:35 +00:00 |
|
|
vasiloglou
|
c441f3e714
|
it is running all bugsfixed
|
2008-02-16 06:24:37 +00:00 |
|
|
Dongryeol Lee
|
90ac6810c9
|
Need to fix the correctness of dense LPR
|
2008-02-16 05:48:51 +00:00 |
|
|
Dongryeol Lee
|
ff573bcc84
|
Little fix on the parameter
|
2008-02-16 02:57:31 +00:00 |
|
|
Dongryeol Lee
|
d3ba206c6b
|
Modified the tree-based code to do train/prediction phase separately
|
2008-02-16 02:24:25 +00:00 |
|
|
rriegel
|
ac83f3b760
|
LOO by index, not dist
|
2008-02-15 22:40:30 +00:00 |
|
|
Dongryeol Lee
|
5c42e89f7b
|
Minor addition to the driver file
|
2008-02-15 22:36:12 +00:00 |
|
|
Dongryeol Lee
|
7bce4938b7
|
Naive LPR confidence band fix
|
2008-02-15 22:12:23 +00:00 |
|
|
Dongryeol Lee
|
449e7c7f13
|
Minior fix to the do LOO nn search
|
2008-02-15 17:59:30 +00:00 |
|
|
vasiloglou
|
e809c275bc
|
Fixed the query=reference case
|
2008-02-15 16:32:55 +00:00 |
|
|
Dongryeol Lee
|
9846258b24
|
Variable bandwidth for naive algorithm implemented
|
2008-02-15 04:26:54 +00:00 |
|
|
Dongryeol Lee
|
104e4ad01a
|
Took out the template argument that specified the local polynomial order
|
2008-02-15 01:50:15 +00:00 |
|
|
Dongryeol Lee
|
b867b7c827
|
Verified the naive algorithm
|
2008-02-15 01:36:27 +00:00 |
|