Word Embeddings

So, I’ve spent the last 2 days working on understanding word embeddings. There are 2 methods I have tried for this,Skip Gram andCBOW


Source code for my attempts: https://github.com/Tzeny/udacity-deep-learning/blob/master/5_word2vec.ipynb

In the git repository above I’ve also included the embeddings matrix resulted from the trained in the form of a .pickle file.

Example predictions

Some examples from the skip-gram training:

 Nearest to one: two, four, seven, eight, three, six, five, nine,  Nearest to man: person, woman, boy, chanute, glasgow, programmatical, trudeau, revenge,  Nearest to concept: form, idea, locational, testimony, result, definition, tordesillas, nisos,