Question for word2vec training

[ThomasDelteil]

It looks like the arguments for args.negative is the number of negative samples, per word context.
I think it would make more sense to use it as the total number of negative samples, as usually discussed in the paper, or update the documentation to make it clear.

            negatives = negatives_sampler(word_context.shape + (args.negative, )) \
                .reshape((word_context.shape[0],
                          word_context.shape[1] * args.negative)) \

Second question:
Is there any mitigation to prevent sampling from word contexts in the negative samples?

@leezu

[ThomasDelteil]

@leezu answered offline:

• args.negative will have the documentation updated
• negative sampling without positive is in PR right now Word embeddings update #159