machine learning - Torch7 alternative to MultiLabelMarginCriterion -


i have multi class label problem solve. ie: each test image can allocated 1-10 non-exclusive labels.

however, having problems multilabelmargincriterion because it's not supported cunn. so, looking alternative approaches. either of these effective?

  1. calculate each permutation of labels in training set (about 150) , train classifier identify 150 classes. however, not think new permutations of labels in test set (those not found in training set) not recognised.

  2. train 10 separate binary classifiers using bcecriterion. ie: 1 classifier each label. run each test image through each classifier , combine results. however, training lots of cnns time-consuming.

the best solution found use msecriterion targets , predictions arrays of 1s , -1s, indicating presence or absence of label.


Comments

Popular posts from this blog

get url and add instance to a model with prefilled foreign key :django admin -

css - Make div keyboard-scrollable in jQuery Mobile? -

ruby on rails - Seeing duplicate requests handled with Unicorn -