probabll / dgm4nlp Goto Github PK

Thanks for the great notebook! Most of the students were able to get through it successfully. There were, however, some common problems that

seminar 1 frequently asked questions:

• a lot of students complained about not being able to test their code right away.
  It might be a good idea to add small code snippets that test each part of the model. Here's an example by Lena
• what is the shape of probs/logits param in bernoulli? ([B])
• are probs probabilities of 0 or 1? (1)
• what is output shape of bernoulli KL? [B]?
• a lot of people got confused with bag-of-words encoder:
  • it wasn't clear what kind of "output dimension" is there. It might be a good idea to clarify that for bag of words case O = I
  • many student asked questions on what VALUES are in the mask. (1's for words that are used, 0's for what is omitted)