Impala Documentation

void Impala::Core::Training::Tester::TestApply ( String kernel ) [inline, private]

note that when we create a reference, this function is dependant on the output of TestTrainModel

Note:: The use of tolerance in tests should not be necessary since in all tests the same data is read. Floating point arithmetic may not be precise but at least it's deterministic right? In any case, this value seems to be small enough that it doesn't matter. The values in the ranking table are around -1e1 so that leaves us 13 significant numbers.

Definition at line 151 of file Tester.h.

References FirstFoldInv(), mDataSrc, mIO, Impala::Core::Training::Svm::Predict(), Impala::Core::Training::TesterIOHelper::ReadModel(), Impala::Core::Training::TesterIOHelper::ReadRanking(), Test(), and Impala::Core::Training::TesterIOHelper::WriteRanking().

Referenced by TestTrainAndApply().

00152     {
00153         FirstFoldInv();
00154         Svm svm;
00155         mIO.ReadModel(&svm, kernel);
00156         Table::ScoreTable* ranking = svm.Predict(mDataSrc);
00157 
00158         mIO.WriteRanking(ranking, kernel);
00159         Table::ScoreTable* refRanking = mIO.ReadRanking(kernel);
00166         double tolerance = 1e-14;
00167         Test(ranking, refRanking, "apply concept " + kernel, tolerance);
00168         delete ranking;
00169         delete refRanking;
00170     }

Here is the call graph for this function: