[jira] [Created] (SPARK-20602) Adding LBFGS as optimizer for LinearSVC

["yuhao yang (JIRA)" <ji...@apache.org>]

yuhao yang created SPARK-20602:
----------------------------------

             Summary: Adding LBFGS as optimizer for LinearSVC
                 Key: SPARK-20602
                 URL: https://issues.apache.org/jira/browse/SPARK-20602
             Project: Spark
          Issue Type: Improvement
          Components: ML
    Affects Versions: 2.2.0
            Reporter: yuhao yang


Currently LinearSVC in Spark only supports OWLQN as the optimizer ( check https://issues.apache.org/jira/browse/SPARK-14709). I made comparison between LBFGS and OWLQN on several public dataset and found LBFGS converges much faster for LinearSVC in most cases.

The following table presents the number of training iterations and f1 score of both optimizers until convergence

||Dataset||LBFGS||OWLQN||
|news20.binary| 31 (0.99) | 413(0.99) |
|mushroom| 28(1.0) | 170(1.0)|
|madelon|143(0.75) | 8129(0.70)|
|breast-cancer-scale| 15(1.0) | 16(1.0)|
|phishing | 329(0.94) | 231(0.94) |
|a1a(adult) | 466 (0.87) | 282 (0.87) |
|a7a | 237 (0.84) | 372(0.84) |

data source: https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html
training code: new LinearSVC().setMaxIter(10000).setTol(1e-6)

LBFGS requires less iterations in most cases (except for a1a) and probably is a better default optimizer. 





--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org