areylng

Results 3 comments of areylng

But you still haven't explained how the optimal threshold for MVSS is determined, why doesn't F1 reach 0.753? Your open-source program also doesn't seem to reach COVER's 0.824. It's strange...

感谢解答,还有一个问题: 在创建测试集的时候,这里以list的方式创建了正负样本。 ![image](https://github.com/bbruceyuan/DeepMatch-Torch/assets/54670610/64220d93-01b7-4637-b638-c1078a345046) 然后推理测试的时候生成了`test_true_label`,value是一个二维的list. ![image](https://github.com/bbruceyuan/DeepMatch-Torch/assets/54670610/00aac10c-21dd-40b9-bc8c-a2bdb96ee1a8) 后面召回了50个电影id,并计算召回率: ![image](https://github.com/bbruceyuan/DeepMatch-Torch/assets/54670610/061adc6f-45a0-4d8e-8f80-a18c4b1e8235) 二维list不能转换成set 这里实际上会报错: ![image](https://github.com/bbruceyuan/DeepMatch-Torch/assets/54670610/123ec3e8-079c-4ba9-9faa-1f35c000d9a3) 此外,这里计算hit的时候,以及计算召回率的时候,是不是都只需要考虑正样本呢? ![image](https://github.com/bbruceyuan/DeepMatch-Torch/assets/54670610/4fe32398-db13-44a1-ba04-35bc4896bea4) ![image](https://github.com/bbruceyuan/DeepMatch-Torch/assets/54670610/ac393676-7420-4983-88c7-599d5ead62f4)

感谢大佬回复 有个题外话,想问下目前大厂做推荐系统是用torch比较多还是用tf.keras比较多呢?