The training is done using the original c code which uses threads to 
paralleize 
(splits the corpus into n parts and each thread process a part in parallel).
The function word2vec in Word2Vec.jl allows you to change the number of 
threads.

But there are rooms for improvement. I noticed a blog post on paralleizing 
word2vec in Python:
http://rare-technologies.com/parallelizing-word2vec-in-python/

It looks promising. I will look into it.

Thanks,

Weijian





On Sunday, 1 November 2015 14:55:14 UTC, Viral Shah wrote:
>
> Are either of these parallel? Any plans to parallelize?
>
> -viral
>
> On Sunday, November 1, 2015 at 6:21:52 PM UTC+5:30, Sergey Bartunov wrote:
>>
>> There is also our package for an extension of word2vec - 
>> https://github.com/sbos/AdaGram.jl which has almost the same speed and 
>> functionality as original word2vec and may additionally learn vectors 
>> corresponding to different meanings of a word. 
>>
>> воскресенье, 1 ноября 2015 г., 12:10:28 UTC+3 пользователь Weijian Zhang 
>> написал:
>>>
>>> Hello,
>>>
>>> We just registered Word2Vec.jl v0.0.1 (
>>> https://github.com/weijianzhang/Word2Vec.jl), a Julia interface to 
>>> word2vec. 
>>> It takes a text corpus as input and produces the word vectors as output.
>>>
>>> You can see a IJulia notebook demo at: 
>>> http://nbviewer.ipython.org/github/weijianzhang/Word2Vec.jl/blob/master/examples/demo.ipynb
>>>
>>> Best wishes,
>>>
>>> Weijian
>>>
>>

Reply via email to