摘要:
ref: http://blog.sina.com.cn/s/blog_8e6f1b330102vu4i.html 阅读全文
摘要:
refer: https://www.cnblogs.com/by-dream/p/9403984.html Out1 = SELECT QueryId, DocId, Rating, ROW_NUMBER() OVER (PARTITION BY QueryId ORDER BY Score DE 阅读全文
摘要:
Architecture SemanticCaps 通过Bi-LSTM表征得到的隐层向量经过multi-head后可以得到R个表征 DetectionCaps 这部分和一般的胶囊网络一样,除了loss function中加入了self attention部分 ... 阅读全文
摘要:
XLNET But the AE language model also has its disadvantages. It uses the[MASK]in thepretraining, but this kind of artificial symbols are absent from the real data atfinetuning time,resulting... 阅读全文
摘要:
A single search -- whether it comes from one of the MSearch items, or a regular single search -- will grab a thread from the search threadpool. The se 阅读全文
摘要:
kd-tree kd-tree works poorly in high dimensions (k<30) 自己实验的时候差不多20到30左右吧,超过之后,就真的很慢了 kd-tree kd-tree works poorly in high dimensions (k<30) 自己实验的时候差不 阅读全文
摘要:
search_k serach_k越大,越准确,但是要在时间和准确率之间取个trade off During the query it will inspect up to search_k nodes which defaults to n_trees * n build on memory or 阅读全文
摘要:
Annoy 随机选择两个点,以这两个节点为初始中心节点,执行聚类数为2的kmeans过程,最终产生收敛后两个聚类中心点 二叉树底层是叶子节点记录原始数据节点,其他中间节点记录的是分割超平面的信息 但是上述描述存在两个问题: (1)查询过程最终落到叶子节点的数据节点数小于 我们需要的Top N相似邻居 阅读全文
摘要:
阅读全文