Skipgram word2vec implementation in C. It implements negative sampling(to improve over vanilla softmax layer as it would be super time-consuming), and uses unigram probability distribution to pick those "negative samples", also uses subsampling(basic probability frequency based) to reduce corpus size.
- text8 dataset
- 1-billion-word-language-modeling-benchmark-r13output dataset