Besides the confusion matrix, how can I get all the misclassification corpus?
I’m looking into Evaluation class, is there any convenient function to do this ?
There is nothing built-in to DL4J that gives you the misclassified examples.
However, it shouldn’t be that hard for you to create your own solution. If you load your data with a record reader, you should be able to set things up to keep metadata so you can tell what the original sample was from the dataset. Then you get the predictions from the model and collect what has been misclassified using that metadata.
Got it.
By the way , is there any build-in classes for multi label evaluation？
You can use model.evaluateROCMultiClass or ROCMultiClass directly, to get quite a nice output when you print out eval.stats(). Just make sure the DataSetIterator knows about your label names 