yoongi0428 / recsys_pytorch Goto Github PK

Seems to be missing files after the V2.0:

• all the generators missing
• data.dataset

Hi,

First of all, thanks for sharing this strong work! I found this repo would be really useful for researchers.

While the overall code structure is very clean and intuitive to me, I have some questions about the evaluation code.

1. I was just wondering what before_evaluation methods(in Model classes) are.

2. I was wondering if your evaluation code is based on the assumption that the shape of the training matrix and test matrix are guaranteed to be the same or not. I found that the numbers of users in training and test matrices might be different as some users are dropped from the matrix in df_to_sparse. Does the evaluation code work even those shapes are not matched?

3. Is there any reference about the input normalization for DAE(CDAE)?

  # normalize
  user_degree = torch.norm(rating_matrix, 2, 1).view(-1, 1)   # user, 1
  item_degree = torch.norm(rating_matrix, 2, 0).view(1, -1)   # 1, item
  normalize = torch.sqrt(user_degree @ item_degree)
  zero_mask = normalize == 0
  normalize = torch.masked_fill(normalize, zero_mask.bool(), 1e-10)

  normalized_rating_matrix = rating_matrix / normalize

Thanks!

Hi yoongi0428,

Thanks a lot for sharing this extensive recommender system implementation.
I have tried to run your code on python 3.8, however I run into trouble with the cython backend.
I get an error of the form:

/local/home/scheidlf/.pyxbld/temp.linux-x86_64-3.8/pyrex/utils/backend/cython/tool.c:610:10: fatal error: numpy/arrayobject.h: No such file or directory
#include "numpy/arrayobject.h"
^~~~~~~~~~~~~~~~~~~~~
compilation terminated.

I was wondering if you could share the environment you worked with or let us know which packages are required to run the code.

Best,
Flo