cjlin1
commented
6 years ago Sorry libsvm doesn't support such incremental settings now. But liblinear does. You can check the paper https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/warm-start/warm-start.pdf to see why it's easier to do linear classification than kernel.
JiayuWu writes:
I mean if dataset is very large, is LibSVM able to train a part first, and then load the model trained and train another part. for example, first time x is [[1, 1], [2, 2]], y is [1, 2], second time x is [[3,3],[4,4]], y is [3,4]. after train two times, I hope the model can classify 4 classes. Now it seems that the model just remember the last train result.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.*
I mean if dataset is very large, is LibSVM able to train a part first, and then load the model trained and train another part. for example, first time x is [[1, 1], [2, 2]], y is [1, 2], second time x is [[3,3],[4,4]], y is [3,4]. after train two times, I hope the model can classify 4 classes. Now it seems that the model just remember the last train result.