| Table::SimilarityTableSet* Impala::Application::LearnConceptFromAnnotations | ( | CmdOptions & | options, | |
| Matrix::DistributedAccess & | da, | |||
| String | concept, | |||
| String | modelname, | |||
| Table::AnnotationTable * | annotations, | |||
| Util::Database * | db | |||
| ) |
Definition at line 25 of file mainActiveLearner.cpp.
References Impala::Util::PropertySet::Add(), Impala::Core::Table::Copy(), Impala::Core::Matrix::DistributedAccess::GetColumnQuids(), Impala::CmdOptions::GetInt(), Impala::Util::PropertySet::GetName(), Impala::Core::Matrix::DistributedAccess::GetRowQuids(), Impala::Util::PropertySet::GetValue(), ILOG_INFO, ILOG_VAR, Impala::Core::Training::Svm::PredictForActiveLearn(), Impala::Core::Training::Svm::SaveModel(), Impala::Core::Table::Select(), Impala::Util::PropertySet::Size(), and Impala::Core::Training::Svm::Train().
Referenced by RunDistributedLearningEngine().
00030 { 00031 ILOG_VAR(Application.LearnConceptFromAnnotations); 00032 00033 00034 ILOG_INFO("concept " << concept); 00035 00036 //load annotation 00037 Table::QuidTable* allQuids = da.GetColumnQuids(); 00038 Table::CriterionElement1InSet<Table::AnnotationTable> c(allQuids); 00039 Table::AnnotationTable* annotation = Select(annotations, c); 00040 00041 Table::QuidTable* positive = annotation->GetPositive(); 00042 Table::QuidTable* negative = annotation->GetNegative(); 00043 Table::QuidTable* learnQuids = positive; 00044 learnQuids->Append(negative); 00045 delete negative; 00046 00047 if((positive->Size() == 0) || (negative->Size() == 0)) 00048 { 00049 delete learnQuids; 00050 delete annotation; 00051 return 0; 00052 } 00053 00054 // read params from best file: the seccond line in the file 00055 /* filename = dataSet->GetFilePathConceptModel 00056 (conceptSet, modelType, featureDef, concept+".best", false, false); 00057 if (filename.empty()) 00058 { 00059 ILOG_INFO("Could not open " << concept << ".best, skipping"); 00060 continue; 00061 }*/ 00062 00063 /* // TODO: estimate .best file 00064 00065 Util::IOBuffer* ioBuf = db->GetIOBuffer(filename, true, false, ""); 00066 String buffer = ioBuf->ReadLine(); 00067 ILOG_INFO(buffer); 00068 buffer = ioBuf->ReadLine(); 00069 Util::PropertySet params(buffer); 00070 */ 00071 Util::PropertySet params; 00072 params.Add("probability", 1); 00073 params.Add("cache", options.GetInt("cache")); 00074 00075 params.Add("gamma", -1); 00076 params.Add("autoweight", 1); 00077 params.Add("kernel", "dist-precomputed"); 00078 params.Add("precompute-kernel", "chi2"); 00079 params.Add("evaluator", "AP"); 00080 params.Add("C", 1.0); 00081 params.Add("w1", 30.0); // estimate these 00082 params.Add("w2", 1.2); // estimate these 00083 for(int i = 0; i < params.Size(); i++) 00084 ILOG_INFO(params.GetName(i) + " " + params.GetValue(i)); 00085 00086 //compute model: 00087 Core::Training::Svm svm; 00088 Core::Training::TrainDataSrcKernelDistributed* dataSrc = new TrainDataSrcKernelDistributed(&da, annotation); 00089 svm.Train(¶ms, dataSrc); 00090 if(!modelname.empty()) 00091 svm.SaveModel(modelname, db); 00092 00093 // compute resulting ranking: 00094 std::vector<std::string> names; 00095 names.push_back(concept); 00096 Table::QuidTable* rowQuids = da.GetRowQuids(); 00097 Table::SimilarityTableSet *simSet = new Table::SimilarityTableSet(names, rowQuids->Size()); 00098 Core::Table::Copy(simSet->GetQuidTable(), rowQuids); 00099 00100 ILOG_INFO("applying model"); 00101 svm.PredictForActiveLearn(da, learnQuids, simSet->GetSimTable(0)); 00102 00103 ILOG_INFO("ranking"); 00104 simSet->ComputeRank(0, true); 00105 00106 delete annotation; 00107 delete learnQuids; 00108 delete dataSrc; 00109 00110 return simSet; 00111 00112 }
Here is the call graph for this function:
