spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "pan (JIRA)" <j...@apache.org>
Subject [jira] [Created] (SPARK-9863) weird output from optimizer limited-memory BFGS
Date Wed, 12 Aug 2015 05:16:47 GMT
pan created SPARK-9863:
--------------------------

             Summary: weird output from optimizer limited-memory BFGS
                 Key: SPARK-9863
                 URL: https://issues.apache.org/jira/browse/SPARK-9863
             Project: Spark
          Issue Type: Question
          Components: MLlib
    Affects Versions: 1.4.1
            Reporter: pan


Hi, I tried to train artificial neural network model using L-BFGS as optimizer but training
process just iterate several times before return and L-BFGS output as following:



5/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 0.7993 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 3.04699 (rel: 0.217) 0.412405 
15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.91066 (rel: 0.0447) 0.172009 
15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.86647 (rel: 0.0152) 0.0880272

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.84330 (rel: 0.00808) 0.0428400

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.83248 (rel: 0.00381) 0.0214614

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82706 (rel: 0.00191) 0.0106791

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82438 (rel: 0.000947) 0.00533878

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82304 (rel: 0.000474) 0.00266689

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82238 (rel: 0.000236) 0.00133327

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82204 (rel: 0.000118) 0.000666502

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82188 (rel: 5.91e-05) 0.000333235

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82179 (rel: 2.95e-05) 0.000166610

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82175 (rel: 1.48e-05) 8.33040e-05

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82173 (rel: 7.38e-06) 4.16516e-05

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82172 (rel: 3.69e-06) 2.08257e-05

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82172 (rel: 1.85e-06) 1.04128e-05

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 9.23e-07) 5.20641e-06

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 4.61e-07) 2.60321e-06

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 2.31e-07) 1.30160e-06

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 1.15e-07) 6.50801e-07

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 5.77e-08) 3.25400e-07

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 2.88e-08) 1.62700e-07

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 1.44e-08) 8.13501e-08

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 7.21e-09) 4.06751e-08

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 3.60e-09) 2.03375e-08

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 1.80e-09) 1.01688e-08

15/08/12 12:24:37 INFO optimize.LBFGS: Step Size: 1.000 
15/08/12 12:24:37 INFO optimize.LBFGS: Val and Grad Norm: 2.82171 (rel: 9.01e-10) 5.08438e-09


What does it mean by the three digital numbers in "INFO optimize.LBFGS: Val and Grad Norm:
2.82171 (rel: 9.01e-10) 5.08438e-09"?  it seems L-BFGS can't move ahead .There is always "2.82171".
Is any one who knows what is problem ?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

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


Mime
View raw message