Github user takuti commented on the issue:
https://github.com/apache/incubator-hivemall/pull/66
**Note on the performance**
For
[news20-multiclass](https://github.com/apache/incubator-hivemall/tree/master/core/src/test/resources/hivemall/classifier)
data, I have translated [our Java test
case](https://github.com/takuti/incubator-hivemall/blob/709848d5626f0df7e7361511224e0e9284b3484d/core/src/test/java/hivemall/topicmodel/OnlineLDAModelTest.java#L147-L223)
to [Python scikit-learn
implementation](https://github.com/takuti-sandbox/tmp/blob/57f740a3d0283e5586cc2cd170a8dd15b9cf96ac/python/lda/news20.py)
w/ (almost) same setting.
In our Java code, unit test finishes in **8 sec** w/ approximately 30
iterations. By contrast, the Python implementation takes around **15 sec** for
30 iterations. Thus, even if `train_lda()` takes very long time for large-scale
data, it should be natural. Hopefully, larger `-delta`, smaller `-iteration` or
smaller `-eps` option could reduce running time (and end up w/ poor results).
* Python code actually creates and handles a 20-by-62061 huge, sparse
matrix. It might be unfair, but Java code alternatively has many inefficient
Map and Array accesses.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
