-
Recently, I am rewriting word2vec implement in c++ version and pyotrch version.
-
Use pytorch to implement word2vec.
-
C++ version cw2vec && word2vec.
-
Now,there are still some problems that need to be improved.
总结就是 两个单词在一句话中,位置距离在2以内, 那么就让他们的embed 后的内积取负号后作为loss 超过2的,让他们的内积加到loss里面 所以bp之后,embed的向量越近的内积大,越远的内积越小.