Difference in model prediction for older vs latest libsvm version

[arunbg]

Hi,

I am trying to update our project's version of Libsvm from version 2.89 to 3.22. I ran into problems on an old test case that was using an SVM model trained with version 2.89. I was able to load the old model using the new 3.22 code, but the predictions vary widely between the versions.

Before I add more details, I wanted to know if old models files are expected to be compatible with the latest code. Since the loading worked, I kind of assumed that the models would be the same. Also, what's the best way to check for equivalence of models?

[cjlin1]

In 3.17 there is a change:

3.17: 2013/04/01 svm.cpp: if class labels are 1 and -1, ensure labels[0] = 1 and labels[1] = -1

this might cause model incompability.

Arun Balagopalan writes:

Hi,

I am trying to update our project's version of Libsvm from version 2.89 to 3.22. I ran into problems on an old test case that was using an SVM model trained with version 2.89. I was able to load the old model using the new 3.22 code, but the predictions vary widely between the versions.

Before I add more details, I wanted to know if old models files are expected to be compatible with the latest code. Since the loading worked, I kind of assumed that the models would be the same. Also, what's the best way to check for equivalence of models?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread. *