wammar
commented
11 years ago - compare likelihood of different models (in particular the latent CRF vs. the HMM)
- compare likelihood of the latent CRF model on the same data with various feature sets
- compare the average reconstruction error for a token (i.e. ranking of the correct token in the ordered list p(z_i|x,z - z_i) )