HaarSingleStumpLearner coredump

[/C=FR/O=CNRS/OU=UMR8607/CN=Francois-David Collin/emailAddress=Francois-David.Collin@…]

{{{multiboost --fileformat arff --csample num 100 --iisize 15x15 --seed 71
--traintest ./uspsHaarTrain.arff ./uspsHaarTest.arff 5 --outputinfo ./OK-
uspsHaar-TreeLearner-HaarSingleStumpLearner-traintest.dta --shypname ./OK-
uspsHaar-TreeLearner-HaarSingleStumpLearner-traintest.xml --learnertype
TreeLearner --baselearnertype HaarSingleStumpLearner 5 --verbose 6
--> Using learner: TreeLearner
Warning: No strong learner is given. Set to default (AdaBoost).
The strong learner is AdaBoostMH
Segmentation fault
}}}

with gdb :

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_INVALID_ADDRESS at address: 0x0000000000000000
0x00000001000e5cf7 in MultiBoost::TreeLearner::initLearningOptions
(this=0x100b00fe0, args=@0x7fff5fbfe600) at
src/WeakLearners/TreeLearner.cpp:90
90                      pWeakHypothesisSource->initLearningOptions(args);

(gdb) l
85                              cerr << "The weak hypothesis must be a
ScalarLearner!!!" << endl;
86                              exit(-1);
87                      }
88
89
90                      pWeakHypothesisSource->initLearningOptions(args);
91
92                      for( int ib = 0; ib < _numBaseLearners; ++ib ) {
93
_baseLearners.push_back(dynamic_cast<ScalarLearner*>(pWeakHypothesisSource->create()));
94
dynamic_cast<BaseLearner*>(_baseLearners[ib])->initLearningOptions(args);

[/C=FR/O=CNRS/OU=UMR8607/CN=Robert Busa-Fekete/emailAddress=busarobi@…]

FD,

please check whether the current version in the SVN has the same problem. I think this bug has been fixed. The problem was in the implementation of function 'cut'.

cheers,robi

[/C=FR/O=CNRS/OU=UMR8607/CN=Francois-David Collin/emailAddress=Francois-David.Collin@…]

it's still crashing :

(gdb) run --fileformat arff --csample num 100 --iisize 15x15 --seed 71 --traintest ./uspsHaarTrain.arff ./uspsHaarTest.arff 5 --outputinfo ./OK-uspsHaar-TreeLearner-HaarSingleStumpLearner-traintest.dta --shypname ./OK-uspsHaar-TreeLearner-HaarSingleStumpLearner-traintest.xml --learnertype TreeLearner --baselearnertype HaarSingleStumpLearner 5 --verbose 6
Starting program: /scratch/fradav/MultiBoostSVN/MultiBoost/multiboost --fileformat arff --csample num 100 --iisize 15x15 --seed 71 --traintest ./uspsHaarTrain.arff ./uspsHaarTest.arff 5 --outputinfo ./OK-uspsHaar-TreeLearner-HaarSingleStumpLearner-traintest.dta --shypname ./OK-uspsHaar-TreeLearner-HaarSingleStumpLearner-traintest.xml --learnertype TreeLearner --baselearnertype HaarSingleStumpLearner 5 --verbose 6
--> Using learner: TreeLearner
Warning: No strong learner is given. Set to default (AdaBoost).
The strong learner is AdaBoostMH

Program received signal SIGSEGV, Segmentation fault.
0x00000000004a9f31 in MultiBoost::TreeLearner::initLearningOptions (this=0x1706ee0, args=@0x7fffaa767da0) at src/WeakLearners/TreeLearner.cpp:90
90                      pWeakHypothesisSource->initLearningOptions(args);
(gdb) backtrace 
#0  0x00000000004a9f31 in MultiBoost::TreeLearner::initLearningOptions (this=0x1706ee0, args=@0x7fffaa767da0)
    at src/WeakLearners/TreeLearner.cpp:90
#1  0x0000000000406ae7 in MultiBoost::AdaBoostMHLearner::run (this=0x172cc10, args=@0x7fffaa767da0)
    at src/StrongLearners/AdaBoostMHLearner.cpp:147
#2  0x00000000004e6ef1 in main (argc=25, argv=0x7fffaa768da8) at src/main.cpp:506
(gdb) 

[anonymous]

The bug was in the registration of HaarSingleStumpLearner because its name in the register did not coincide with the name of the class. So, we had changed this line

REGISTER_LEARNER_NAME(HaarSingleStump, HaarSingleStumpLearner)

to

REGISTER_LEARNER(HaarSingleStumpLearner)

The shouldn't use REGISTER_LEARNER_NAME macro later on.