Timeline for Is a high number of support vectors a sign of overfitting?
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 9, 2015 at 20:04 | answer | added | Marc Claesen | timeline score: 3 | |
| Sep 9, 2015 at 18:55 | comment | added | yi.tang.uni | @fcoppens I changed the title based on your suggestion. thanks. | |
| Sep 9, 2015 at 18:53 | history | edited | yi.tang.uni | CC BY-SA 3.0 |
edited title
|
| Aug 31, 2015 at 16:00 | comment | added | user83346 | I think you have a problem of overfitting, there are some results on error bouds and the lnk with the number of support vectors but I am not an expert in that. Nevertheless a very high fraction of support vectors is an indication of overfit. Maybe you should re-phrase your question "Is a high number of support vectors a sign of overfitting ? " | |
| Aug 31, 2015 at 15:27 | comment | added | yi.tang.uni | @fcoppens there are 196 in training set and 196 in test set, both are derived from random splitting the full dataset. | |
| Aug 31, 2015 at 15:26 | comment | added | yi.tang.uni | @user777 that's true! I randomly split the observation set into training and test set with 50/50 split. I guess I didn't the expect the relationships between x and y are so different among them. | |
| Aug 31, 2015 at 15:23 | comment | added | user83346 | How many observations did you have in the training sample ? | |
| Aug 31, 2015 at 15:20 | comment | added | Sycorax♦ | But this is a different data set, correct? Perhaps the relationship between features and outcome is weaker in this data, or there are many low-quality input features, or you need a more specialized kernel function, or you need more data to learn the boundary in the feature space... | |
| Aug 31, 2015 at 15:18 | comment | added | yi.tang.uni | @user777 by the difference between training and test error, that is about 28%. I've applied the SVM to other dataset, the differences are usually about 5-10%. | |
| Aug 31, 2015 at 15:14 | review | First posts | |||
| Aug 31, 2015 at 15:20 | |||||
| Aug 31, 2015 at 15:13 | comment | added | Sycorax♦ | Test error will always be larger than training error. Why are you surprised, precisely? | |
| Aug 31, 2015 at 15:10 | history | asked | yi.tang.uni | CC BY-SA 3.0 |