Coverage for hopwise/model/general_recommender/pop.py: 86%

1# @Time : 2020/8/11 9:57 

2# @Author : Zihan Lin 

3# @Email : linzihan.super@foxmail.com 

4# UPDATE 

5# @Time : 2020/11/9 

6# @Author : Zihan Lin 

7# @Email : zhlin@ruc.edu.cn 

8# UPDATE 

9# @Time :2023/9/21 

10# @Author : Kesha Ou 

11# @Email :1582706091@qq.com 

12 

13r"""Pop 

14################################################ 

15 

16""" 

17 

18import torch 

19 

20from hopwise.model.abstract_recommender import GeneralRecommender 

21from hopwise.utils import InputType, ModelType 

22 

23 

24class Pop(GeneralRecommender): 

25 r"""Pop is an fundamental model that always recommend the most popular item.""" 

26 

27 input_type = InputType.POINTWISE 

28 type = ModelType.TRADITIONAL 

29 

30 def __init__(self, config, dataset): 

31 super().__init__(config, dataset) 

32 

33 self.item_cnt = torch.zeros(self.n_items, 1, dtype=torch.long, device=self.device, requires_grad=False) 

34 self.max_cnt = None 

35 self.fake_loss = torch.nn.Parameter(torch.zeros(1)) 

36 self.other_parameter_name = ["item_cnt", "max_cnt"] 

37 

38 def forward(self): 

39 pass 

40 

41 def calculate_loss(self, interaction): 

42 item = interaction[self.ITEM_ID] 

43 self.item_cnt[item, :] = self.item_cnt[item, :] + 1 

44 

45 self.max_cnt = torch.max(self.item_cnt, dim=0)[0] 

46 

47 return torch.nn.Parameter(torch.zeros(1)).to(self.device) 

48 

49 def predict(self, interaction): 

50 item = interaction[self.ITEM_ID] 

51 result = torch.true_divide(self.item_cnt[item, :], self.max_cnt) 

52 return result.squeeze(-1) 

53 

54 def full_sort_predict(self, interaction): 

55 batch_user_num = interaction[self.USER_ID].shape[0] 

56 result = self.item_cnt.to(torch.float64) / self.max_cnt.to(torch.float64) 

57 result = torch.repeat_interleave(result.unsqueeze(0), batch_user_num, dim=0) 

58 return result.view(-1)