Quote:
Originally Posted by a.sanyal902
I've tried several runs in this question. Howver, I still get the lowest cross-validation error for C=0.1, rather than 0.001. Their errors are pretty close though. I did 100 runs for each of the C's and then took an average value of the errors.
I'd like to know what everyone else has done in the question and whether they got results similar to mine. Would it make a difference if I compared the errors after 1 run with each of the C's and then, finally selected the C which gets selected most often?
P.S I randomized the seed of svm-train.c, as suggested here. |
I tried to compare the errors after 1 run and nothing worked. Perhaps, you have found the solution already. Would appreciate any ideas.