SVM: one_vs_rest error

Asked Mar 01 '15 at 02:27

Active Mar 01 '15 at 02:33

Viewed 64 times

I use libsvm to do SVM Classification and I use one-over-rest method. The problem is when I run the following command, I found the values for "accuracy" are not all real numbers, there is NaN in it. Can anyone tell me how to fix the problem? Thank you so much!

model = ovrtrainBot(trainLabel, trainData, bestParam);

[predict_label, accuracy, decis_values] = ovrpredictBot(testLabel, testData, model);

There are strange values in "accuracy":

accuracy=[92.56 93.39   97.521  92.562  91.74 92.562    93.39   93.39 93.39
         0.074  0.07     0.025  0.074   0.083   0.074   0.07    0.066   0.066
         0.73   0.13     0.65   NaN     0.0007  NaN     0.104   0.433   0.104]

edited Mar 01 '15 at 02:33

asked Mar 01 '15 at 02:27

Angelababy

http://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html#f427 – rayryeng Mar 01 '15 at 19:33
@rayryeng Thank you so much! I am trying to fix the problem! – Angelababy Mar 01 '15 at 20:24
@rayryeng, in my computer, I do not have VPN, I never set up one. It seems it is not this problem. – Angelababy Mar 02 '15 at 02:26
@rayryeng, even if I disconnect the entire network, the result still has "NaN", I do not think VPN is the reason to cause NaN. – Angelababy Mar 02 '15 at 21:14
Hmm... I'm not sure. I'll have to look into it for you – rayryeng Mar 02 '15 at 21:15
@rayryeng, my dataset was z-scored before I put it into SVM. Do you think z-scoring the dataset is the reason to cause the value "NaN"? – Angelababy Mar 02 '15 at 21:17
@rayryeng, I send you an email. Please take a look at it. Thank you! Waiting for your response. – Angelababy Mar 04 '15 at 02:00

SVM: one_vs_rest error

0 Answers0