fangleai / implicit-lvm Goto Github PK

fasttext.supervised in line 150 on train_yelp.py not working anymore. Please may I know what version of fasttext has to be installed to run it?

Hi, I am trying to reproduce the results of this paper, more specifically the iVAE model for the yahoo language modeling dataset. I wanted to know about the loss behavior during training. My model is currently on the 16th epoch and the negative ELBO is still 526, whereas in the paper it is 309 after 40 epochs of training. I am also not seeing a decreasing trend in my elbo and reconstruction loss. Is it supposed to drop drastically later on as we train or this is unexpected? I have only followed the instructions provided in the repository to train the model.
Thank you

Hi,
Where is the script for test phase? i.e., BLEU, dist, BOW for dialogue response generation task.

Hi there,

I scanned the code of style transfer in train_yelp.py and arae_train_yelp.py. I found that they are quite similar. I am not sure whether train_yelp.py uses the implicit vae with mutual information for training because the Nu_z function might not be included.

Could you please give me some suggestions about it? I might miss some information in the style transfer folder.

Many thanks~