复现UPC-SDG

复现Privacy-Preserving Synthetic Data Generation for Recommendation Systems的UPC-SDG遇到了一点问题

记录自己的复现流程

1. 首先按照GitHub所说：
   （1）Create the empty folders, output and data.
   （2）Download the train data from the Amazon Review Data and SNAP Page, details setting see Dataset section
   （3）Prepare for pre-trained User/Item embedding weight from Google Ddrive and put them in ./code/embedding

2. 数据准备好之后会出现路径问题
   在电脑上的话，要修改的代码是main.py line28
   pretrain_file_path = ‘./embedding/mf-Office-64.pth.tar’ #原’./code/embedding/mf-Office-64.pth.tar’
   还有
   \Procedure.py", line 219返回值对不上，需要：
   need_replace, replaceable_items, replaceable_items_feature, feature_loss = \
   recommend_model.computer_pos_score(unique_user, user_pos_items, mask, train_pos)
   改成
   need_replace, replaceable_items, replaceable_items_feature, similarity_loss, similarity, feature_loss = \
   recommend_model.computer_pos_score(unique_user, user_pos_items, mask, train_pos)

3. 代码执行时
   需要先进入D:\Code\UPC-SDG-main\UPC-SDG-main\code
   然后执行代码 执行时去掉原作者的./code
   直接是
   python -u main.py --decay=1e-1 --lr=0.001 --seed=2022 --dataset=“Office” --topks=“[20]” --recdim=64 --bpr_batch=2048 --load=1 --replace_ratio=0.2 --privacy_ratio=0.1 --bpr_loss_d=1 --similarity_loss_d=3

大功告成~
不不不 想要图形化界面运行
还需要用pycharm直接到code目录 然后改parse.py line42
parser.add_argument(‘–privacy_ratio’, type=float, default=0.1,
help=‘set the ratio of user’s privacy sensitivity’)