Timeline for Why does SVM needs to keep support vectors?
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 1, 2016 at 20:44 | vote | accept | Octoplus | ||
| Jun 1, 2016 at 9:38 | answer | added | dontloo | timeline score: 7 | |
| Jun 1, 2016 at 8:44 | comment | added | Dikran Marsupial | "they often end up retaining only a small fraction of the number of examples." your mileage may vary on this one, in my experience if you choose the hyper-parameters to maximise generalisation performance you quite often end up with virtually all the data being support vectors. This depends on the nature of the problem and the hyper-parameter settings. | |
| Jun 1, 2016 at 8:18 | answer | added | Daneel Olivaw | timeline score: 4 | |
| May 31, 2016 at 23:33 | comment | added | A. Ray | I think what it is trying to say is the optimal $w$ is dependent on just the small number of support vectors. You are correct that once $w$ is computed you don't really need to save these support vectors if you just need to classify. However if new data points arrive in the future, you may have to store all of them and recompute $w$ every time with new data. | |
| May 31, 2016 at 22:08 | history | asked | Octoplus | CC BY-SA 3.0 |