Doc2Vec docstring updated to reflect actual default value of sample parameter. Fix #1302. #1307
Conversation
tmylk
changed the title
| @@ -580,7 +580,7 @@ def __init__(self, documents=None, dm_mean=None, | |||
| need about 1GB of RAM. Set to `None` for no limit (default). | |||
|
|
|||
| `sample` = threshold for configuring which higher-frequency words are randomly downsampled; | |||
| default is 0 (off), useful value is 1e-5. | |||
| default is 1e-3, useful value is 1e-5. | |||
There was a problem hiding this comment.
Thanks for the fix.
This doc reads very strange now: default value is X, useful value is Y. Why not use the useful value? 🤔
@gojomo
There was a problem hiding this comment.
I think the 'useful value' remark dates back to something in word2vec.c code, before it had a (modest) non-zero default. I suppose we could stop at just mentioning the default, or hint that more aggressive values are also worth trying with a comment like: "default is 1e-3; values to 1e-5 (or lower) may also be useful". (On large datasets, I've seen ok results with 1e-06 and 1e-07, which each offer additional training speedups.)
There was a problem hiding this comment.
Agreed. @menshikh-iv @datapythonista can you do this tiny fix?
There was a problem hiding this comment.
@piskvorky Yes, of course.
No description provided.