Baffling: Very long runtime on thor compared to hthor

[anirudh]

Greetings!

I have a small piece of ECL code that is taking 1.7 seconds to run on hthor while it take 38 mins to run on thor cluster (10 node cluster).

The log files are available at: https://www.dropbox.com/sh/yvib442myp3yih1/AAAnYpE13fe9OXS9Vs_xer_ea?dl=0

Any pointers would be very helpful.

Code: Select all

IMPORT ML;

value_record := RECORD
      unsigned    rid;
      real       age;
      real       height;
      integer1    sex; // 0 = female, 1 = male
END;
         
d := DATASET([{1,35,149,0},{2,11,138,0},{3,12,148,1},{4,16,156,0},
              {5,32,152,0},{6,16,157,0},{7,14,165,0},{8,8,152,1},
               {9,35,177,0},{10,33,158,1},{11,40,166,0},{12,28,165,0},   
               {13,23,160,0},{14,52,178,1},{15,46,169,0},{16,29,173,1},
               {17,30,172,0},{18,21,163,0},{19,21,164,0},{20,20,189,1},
               {21,34,182,1},{22,43,184,1},{23,35,174,1},{24,39,177,1},
               {25,43,183,1},{26,37,175,1},{27,32,173,1},{28,24,173,1},
               {29,20,162,0},{30,25,180,1},{31,22,173,1},{32,25,171,1}]
              ,value_record);
                                                                                                               
ML.ToField(d,flds0);
f4 := PROJECT(flds0(Number=3),TRANSFORM(ML.Types.NumericField,SELF.Number := 4,SELF.Value := 1-LEFT.Value,SELF := LEFT));
flds1 := flds0+f4;
flds := ML.Discretize.ByRounding(flds1);
LogisticModule := ML.Classify.Logistic();

Model3 := LogisticModule.LearnCS(flds0(Number<=2),flds(Number=3));
Model3;
Model4 := LogisticModule.LearnCS(flds0(Number<=2),flds(Number=4));
Model4;

TestModule := LogisticModule.TestD(flds(Number<=2),flds(Number>=3));
TestModule.CrossAssignments;
TestModule.PrecisionByClass;
TestModule.Accuracy;

LogisticModule.ClassifyC(flds0(Number<=2),Model3);
LogisticModule.ClassifyC(flds0(Number<=2),Model4);