Add "most_similar_to_given" method for KeyedVectors #1582
Conversation
.spyproject/codestyle.ini
Outdated
| @@ -0,0 +1,6 @@ | |||
| [codestyle] | |||
There was a problem hiding this comment.
What is it? Please remove all non-relevant files (all from .spyproject folder).
gensim/models/keyedvectors.py
Outdated
| @@ -617,6 +618,22 @@ def similarity(self, w1, w2): | |||
|
|
|||
| """ | |||
| return dot(matutils.unitvec(self[w1]), matutils.unitvec(self[w2])) | |||
|
|
|||
| def most_similar_to_given(self, w1, word_list): | |||
There was a problem hiding this comment.
Please reformat your docstring according to google-style
gensim/models/keyedvectors.py
Outdated
|
|
||
| Example:: | ||
|
|
||
| >>> trained_model.most_similar_to_given('music', ['water', 'sound', 'backpack', 'mouse']) |
There was a problem hiding this comment.
What do you this @gojomo, it's a useful feature?
There was a problem hiding this comment.
If done efficiently it makes sense. If the common case is there's a stable subset against which some projects are doing all of their similarity-searches, the need might be best met by a new method for creating a subset KeyedVectors, with just the words-of-interest. Overlaps with closed-as-idle PR #1229.
menshikh-iv
changed the title
menshikh-iv
changed the title
|
@TheMathMajor Also, please fix PEP8 issues (look at travis log) |
|
There's another nearly-complete implementation of similar functionality by @shubhvachher in closed-as-idle PR #1229. |
|
Ping @TheMathMajor, what's a status of this PR? |
|
Hi, thanks for the feedback, I have made the committed the changes requested. |
|
Thanks @TheMathMajor LGTM |
|
There's no need for "deprecated" forwarding-method in Word2Vec if this is a brand-new feature on Perhaps the method should have a test, but as a simple 1-liner composed of other well-tested methods, maybe not. But that highlights another difference with the earlier #1229 – while that PR had a lot of code-duplication, it did try to do the similarity calculations with array math, and thus might be noticeably faster with long word-lists. If main goal is performance, that approach may have been better; if goal is simply providing a convenience/clarity/example-method, this idiomatic 1-liner is better. |
Agree, I'll remove it from Word2Vec For this method, I think clean one-liner is better (IMO we no need performance here) |
|
Thanks @TheMathMajor, congratz with the first contribution:1st_place_medal: |
|
Thanks a lot for the suggestions and guiding me through my first contribution! |
* finished adding 2 new functions * imported argmax to word2vec * reformatted * remove `most_similar_to_given` from w2v class * Fix PEP8
Added a function to find the most similar word in a given list to a given word.