Closed 3NFBAGDU closed 4 years ago
not sure but here's a good source that explains all the loss functions that they have used: https://www.cis.uni-muenchen.de/esslli2017/pdf/print,embedgd.pdf
thanks, this is a good reference. the question is the same for word2vec and sent2vec. it's hard to say which loss level would ensure good quality, as it depends on many unknowns and other factors of the training corpus. closing this now
What is the acceptable loss for sent2vec after training?