skip-gram-pytorch
skip-gram-pytorch copied to clipboard
A complete pytorch implementation of skip-gram
Why do i feel that the loss isnt decreasing? It started at 13 and then jumped to 8 and now i see random losses like 7 or 6 or 5...
How did you come up with equation for subsampling selection probability?
it seems that random text is not suitable for running the code, could you please share the text8 file?
I started off from your code and made a few modifications described in detail on this blog: http://www.telesens.co/2018/02/08/word2vec-experiments/ The major modification was adding the ability to use a pre-trained model...