Ngram detection Phrases vs Phraser

[Simon Lindgren]

Gensim is great!

I have this code for detecting bigrams:

# Find bigrams in the texts
texts2 =[]
from gensim.models import Phrases
bigram = Phrases(texts) # find bigrams
for t in texts:
 texts2.append(bigram[t]) # append the bigrams version of the text

It works! But it throws this error:

/python/site-packages/gensim/models/phrases.py:274: UserWarning: For a faster implementation, use the gensim.models.phrases.Phraser class
  warnings.warn("For a faster implementation, use the gensim.models.phrases.Phraser class")

I find no Phraser to import in Gensim. Should I do things differently?

[Shubh Vachher]

Check this out. https://radimrehurek.com/gensim/models/phrases.html

[Gordon Mohr]

That's a warning rather than an error - so if everything else seems to be working, there's no need to change anything. 

The warning is kind of misleading, too. `Phraser` is an alternative that must be built from an initial `Phrases` instance, in a time-consuming step. It then works a little faster but uses much less memory – but that Phraser is stuck with the min_count/threshold parameters in effect when it was constructed. In contrast, if you retain the larger/slower `Phrases`, you can still tinker with those parameters. So `Phraser` is not really a drop-in 'faster implementation' everyone would want to consider. You'd need to review the doc-comments & code to evaluate if it's helpful for your needs.

- Gordon