Closed tctctctctc closed 2 years ago
感谢关注,但是如我们在 #485 中声明的:
Note, although we have tested all the implemented models on at least two datasets (most are tested on four datasets), we will never maintain a scoreboard to compare the performance of different algorithms. In other words, we will not release any test results of the implemented models. However, we have prepared the default parameter settings for all models. You can simply obtain their performance with very simple commands using the default parameters.
目前我们不会发布任何测试结果。
更多细节请参考 #485,希望理解。
感谢,不知后续是否会推出更加详尽的中文手册,对重要配置参数做更详细的介绍?
感谢反馈,团队之前有考虑过增加中文 API 文档,但是由于人手问题,暂未安排相关计划。
希望未来能有机会,与社区一同完善双语文档。
请问序列模型的数据集只能按留一法划分嘛
请问序列模型的数据集只能按留一法划分嘛
RecBole 1.0.0 为序列模型提供了留一法、按比例划分等方式,详见 APC Doc。
已有工作中,有一些采用留一法,如 S^3-Rec [1];也有些采用按比例划分,如 session-based 设置下的 ProxySR [2]。
By the way,新的主题的讨论建议新开 discussion / issue。
Ref: [1] S^3-Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization. CIKM 20. [2] Unsupervised Proxy Selection for Session-based Recommender Systems. SIGIR 21.
Closing due to inactivity. Please comment if you're still having issues.
既然提供了运行ml-1m,Netflix和yelp三个数据集的建议配置和时间花费,能否把所有模型在以上数据集运行后的最佳结果和相应参数也展示一下