René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Three questions","articleBody":"Thank you for your work and I have three questions:\r\n1. Why do you use this method to calculate the initialization range? I didn't see the relevant introduction in your paper. What's the purpose of this method?\r\n\r\nself.embedding_range = nn.Parameter(\r\n    torch.Tensor([(self.gamma.item() + self.epsilon) / hidden_dim]),  \r\n    requires_grad=False\r\n)\r\n\r\nself.entity_embedding = nn.Parameter(torch.zeros(nentity, self.entity_dim))\r\nnn.init.uniform_(\r\n    tensor=self.entity_embedding, \r\n    a=-self.embedding_range.item(), \r\n    b=self.embedding_range.item()\r\n)\r\n\r\n2. This range is also used when pluralizing relationships. Why can this be done？\r\n\r\nphase_relation = relation/(self.embedding_range.item()/pi) \r\nre_relation = torch.cos(phase_relation) \r\nim_relation = torch.sin(phase_relation) \r\n\r\n3. In the rotate model, the calculations of head batch  and tail batch are different in sign, but in the paper i can't find the head-batch part, i can't understand this part\r\n\r\nif mode == 'head-batch':\r\n    re_score = re_relation * re_tail **+** im_relation * im_tail\r\n    im_score = re_relation * im_tail **-** im_relation * re_tail\r\n    re_score = re_score - re_head\r\n    im_score = im_score - im_head\r\nelse:\r\n    re_score = re_head * re_relation **-** im_head * im_relation\r\n    im_score = re_head * im_relation **+** im_head * re_relation\r\n    re_score = re_score - re_tail\r\n    im_score = im_score - im_tail\r\n ","author":{"url":"https://github.com/zulihit","@type":"Person","name":"zulihit"},"datePublished":"2022-05-26T07:38:12.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/55/KnowledgeGraphEmbedding/issues/55"}