Experimental code and results of:
T. Kitazawa, T. Matsuo, Incremental Approaches for Matrix Approximation: Performance Evaluations and Their Possible Applications, (in Japanese), The Japanese Society for Artificial Intelligence SIG-FPAI-B501, Aug. 2015.
Training data of USPS normalized handwritten dataset.
http://statweb.stanford.edu/~tibs/ElemStatLearn/data.html
Running Time: around 4000 sec.
Running Time: around 750 sec.
| k | Covariance Error | Projection Error |
|---|---|---|
| 2 | 0.0657172694621 | 1.0 |
| 4 | 0.0491418902379 | 1.0 |
| 8 | 0.0256682270686 | 1.0 |
| 16 | 0.0122495458957 | 1.0 |
| 32 | 0.0052127371377 | 1.0 |
| 64 | 0.00168276318593 | 1.0 |
| 128 | 0.000446738422159 | 1.0 |
| 256 | 6.82607484728e-13 | 5228.83140371 |
| ell | Projection Error | Running Time (sec.) |
|---|---|---|
| 4 | 1.9043975495 | 0.591689 |
| 8 | 1.53392089235 | 0.877754 |
| 16 | 1.14943840055 | 2.401179 |
| 32 | 1.00044963427 | 6.108395 |
| 64 | 1.00000728168 | 17.863258 |
| 128 | 1.00000006729 | 84.104562 |
| 256 | 1.0 | 304.967199 |