search

HuilinChenJN / UPC_SDG

SIGIR 2022: Privacy-Preserving Synthetic Data Generation for Recommendation Systems
8 stars 3 forks source link

CF2_loss calculation seems to be incorrect? #2

Open littlemilkwu opened 1 year ago

littlemilkwu commented 1 year ago

https://github.com/HuilinChenJN/UPC_SDG/blob/711ea4b8fd8ae93ed3d83cbdea867db4d934a970/code/model.py#L216

I compared your code with your paper, and I found that equation (14) is linked to line 216 in model.py. In your code, you replaced the combination of the sigmoid function and ln function with the softplus function. However, it seems that they are not exactly the same: softplus(x) = ln(1 + sigmoid(x))

Additionally, I'm not sure why you used "mean" instead of "sum" in the same line. Please correct me if I'm mistaken. I would greatly appreciate it!

DoJerryChen commented 1 year ago

File "./code/main.py", line 81, in Procedure.output_generative_data(dataset, Recmodel, weight_file) File "UPC_SDG-main\code\Procedure.py", line 220, in output_generative_data recommend_model.computer_pos_score(unique_user, user_pos_items, mask, train_pos) ValueError: too many values to unpack (expected 4)

Have you encountered this problem? When running the first office program, it reported an error on epoch 140.