jingsenzhang / recbole-debias Goto Github PK

Hello, first of all, thank you to your team for developing rebole.

During the process of using rebole-debias, I followed your config in the Details and downloaded the data you provided in Here.

I ran the MF model and MF_IPS model separately on this dataset with corresponding configurations. But The MF_IPS model shows worse than MF.

May I ask if this is a normal phenomenon? If not, how can I get better performace of MF_IPS model than MF model?

Looking forward to your reply.
Thanks.

您好，贵组的Recbole-Debias的代码config文件里的configurator.py的第70行：self.internal_config_dict = dict()
似乎不需要重新创建internal_config_dict
而且def _load_internal_config_dict这个方法是不是应该加一行：super()._load_internal_config_dict(model, model_class, dataset)
来加载父类Recbole里的配置信息
也就是说需要加载父类Recbole里的配置信息，而不需要重新创建一个字典
如有错误，请指教，谢谢！

In many debiasing papers (DICE, MACR) it is mentioned that the test set was created by sampling items uniformly.
Can you tell me what part of the config file I need to modify if I want to sample the test set in the following way?
It seems like you only have 'train_neg_sample_args' in your DICE config file, which is about sampling the negatives during the training process, not the test data.

Hi, thank you for your work. Now I tried to run your project in a new conda env. I installed the latest recbole. But the error occurs. Do I need to correct any recbole versions?

I use the code provided by Recbole's official website:

from recbole_debias.quick_start import run_recbole_debias
run_recbole_debias(model='PDA', dataset='ml-100k')

However, I got the following error:

AttributeError: Metric 'custommetric' has no attribute [metric_need].

How can I use this package?

Don't DICE, MACR, PDA have different implementations of MF and LightGCN? Why is it not reflected in this code

使用的模型是DICE，验证方式采用了leave one out的方式，repeatable = True 和 repeatable = False的结果是一样的，想问一下目前是否还没有repeatable这一功能？

ValueError: Usecols do not match columns, columns expected but not found: ['user_id:token', 'timestamp:float', 'rating:float', 'item_id:token']