machine learning - Late fusion step of classification using libLinear -
i doing classification work use liblinear kernel these days. , have trained 2 type of feature sets 2 models prediction query input. wish utilize late fusion combine 2 result models, change code of liblinear can decision score different classes. got 2 sets of score determine class query should in.
is there standard way "late fusion" or intuitively add 2 scores of each classes , choose class highest score candidate?
the standard way combine multiple classifiers weighted sum of scores of individual classifiers. of course, have problem of specifying weight coefficients. there different possibilities:
- set weights uniformly
- set weights proportional performance of classifier
- train new classifier takes scores input
Comments
Post a Comment