Integrate softmax cpr to the latest RecBole version #1912
Conversation
|
|
||
| all_memory_W = self.dropout(self.We(all_memory)) | ||
| all_memory_W = all_memory_U.unsqueeze(1) | ||
| all_memory_W = all_memory_U.expand(bsz, all_memory.size(1), seq_len, hsz) |
There was a problem hiding this comment.
all_memory_U here should be replaced by all_memory_W
| bsz, seq_len_2 = item_seq.size() | ||
| logit_hidden_context = (projected_emb_arr[self.n_facet + self.n_facet_reranker*len(self.reranker_CAN_NUM) + i].unsqueeze(dim=2).expand(-1,-1,seq_len_2,-1) * test_item_emb[item_seq, :].unsqueeze(dim=1).expand(-1,seq_len_1,-1,-1) ).sum(dim=-1) | ||
| if test_item_bias is not None: | ||
| logit_hidden_reranker_topn += test_item_bias[item_seq].unsqueeze(dim=1).expand(-1,seq_len_1,-1) |
There was a problem hiding this comment.
logit_hidden_reranker_topn here should be logit_hidden_context
|
@ken77921 Nice job! I've pointed out some minor mistakes in the code. Besides, you can add your personal information in the header of the model files. And it would be highly appreciated if you can complete the comments of the new parameters in the two yaml files so that the users can understand their meaning quickly. Last but not least, the test functions for these two models need to be added to the file |
|
@BishopLiu Thanks for the corrections. I am not very sure how to test |
|
@ken77921 Thank you for your great contribution! I'll merge it. |
|
Thank you for merging the pull request. I have a question. Will the descriptions of GRU4RecCPR and SASRecCPR be added to the documentation (https://recbole.io/docs/user_guide/model_intro.html)? |
|
Yes, we will add it during the next version update. |
The code is modified from https://github.com/iesl/softmax_CPR_recommend