%------ Leave-one-out ----------------- sepPosNeg([],[],[]). sepPosNeg([TrainEx|TrainingEIs],[TrainEx|TrainPE],TrainNE):- ex(TrainEx,1), sepPosNeg(TrainingEIs,TrainPE,TrainNE). sepPosNeg([TrainEx|TrainingEIs],TrainPE,[TrainEx|TrainNE]):- ex(TrainEx,0), sepPosNeg(TrainingEIs,TrainPE,TrainNE). leave_one_out(FoldID,HSampleSize):- findall(Ex,ex(Ex,PosNegSign),AllExs), nth(FoldID,AllExs,TestEx), delete(AllExs,TestEx,TrainExs), sepPosNeg(TrainExs,TrainPos,TrainNeg), statistics(cputime,[Total1,Previous]), generalise_oneTrainingSetF(TrainPos,TrainNeg,HSampleSize,Hs), statistics(cputime,[Total2,TimeTaken]), remove_duplicates(Hs,Hs_noDuplicate), tell('h_metaBayes.pl'),portray_clause(sampledHs(Hs_noDuplicate)),told, % tell('h_metaBayes_superimposed.pl'), % superimpose_clauses(Hs_noDuplicate,ClausesWithFrequencists), % told, ex(TestEx,TargetLabel), get_acc_likelihood(Hs_noDuplicate,TestEx-TargetLabel,PA,LogLikelihood), length(Hs_noDuplicate,N),AveTimeTaken is TimeTaken/(N*1000), portray_clause(numSampledHs(N)), tell('onePA.pl'), write(PA),write(' '), told, tell('oneLogLikehood.pl'), write(LogLikelihood),write(' '), told, tell('oneTime.pl'), write(TimeTaken),write(' '), told, write(result_oneSample(Problem_info,SampleSize,SameSizeRepeated,PA,LogLikelihood,AveTimeTaken)),writeln('.'),nl,!. % ------ hold out: assuming training examples are in the form of ex(Ex,PosNegSign), while test examples are as test_examples(Ex,PosNegSign). train_test(HSampleSize):- findall(Ex,ex(Ex,1),TrainPos), findall(Ex,ex(Ex,0),TrainNeg), write('%'),write({TrainPos,TrainNeg}),nl, statistics(cputime,[Total1,Previous]), %write('++++++++'),writeln(PLKB), %nth1(1,PLKB,PL),nth1(2,PLKB,KB), %write('++++++++'),writeln(PL), %write('++++++++'),writeln(KB), generalise_oneTrainingSetF(TrainPos,TrainNeg,HSampleSize,Hs), statistics(cputime,[Total2,TimeTaken]), remove_duplicates(Hs,Hs_noDuplicate), tell('hypothesises.pl'),portray_clause(sampledHs(Hs_noDuplicate)),told, findall(TestEx-TargetLabel,test_examples(TestEx,TargetLabel),AllTest_withLabels), maplist(get_acc_likelihood(Hs_noDuplicate),AllTest_withLabels,PAs,LogLikelihoods), sumList(PAs,PA0),length(AllTest_withLabels,NumTestExs),PA is 100*PA0/NumTestExs, sumList(LogLikelihoods,LogLikelihood), length(Hs_noDuplicate,N),AveTimeTaken is TimeTaken/(N*1000), portray_clause(numSampledHs(N)), tell('onePA.pl'), write(PA),write(' '), told, tell('oneLogLikehood.pl'), write(LogLikelihood),write(' '), told, tell('oneTime.pl'), write(AveTimeTaken),write(' '), told. train_testF(SampleSize,SameSizeRepeated,HSampleSize):- write('%%%%%train_testF'),write({SampleSize,SameSizeRepeated}),nl, oneSampleF(SampleSize,SameSizeRepeated,TrainPos,TrainNeg), write('%'),write({TrainPos,TrainNeg}),nl, statistics(cputime,[Total1,Previous]), %write('++++++++'),writeln(PLKB), %nth1(1,PLKB,PL),nth1(2,PLKB,KB), %write('++++++++'),writeln(PL), %write('++++++++'),writeln(KB), generalise_oneTrainingSetF(TrainPos,TrainNeg,HSampleSize,Hs), statistics(cputime,[Total2,TimeTaken]), remove_duplicates(Hs,Hs_noDuplicate), tell('h_metaBayes.pl'),portray_clause(sampledHs(Hs_noDuplicate)),told, tell('h_metaBayes_superimposed.pl'), superimpose_clauses(Hs_noDuplicate,ClausesWithFrequencists), told, /* group_bySize(Hs0,GroupedHs), renormalise_blocks(GroupedHs,Hs), portray_clause(updated(Hs)), dataList(test,pos,TestPos),dataList(test,neg,TestNeg), write('\n%%%%% Bayes prediction results: '),nl, unCoverRecord_block(Hs,TestPos,FN),length(FN,FalseNegativeNum), write(false_negative(SampleSize,SameSizeRepeated,FN)),writeln('.'),% False Negative unCoverRecord_block(Hs,TestNeg,TN),length(TN,TrueNegativeNum), write(true_negative(SampleSize,SameSizeRepeated,TN)),writeln('.'),nl,% True Negative */ dataList(test,pos,TestPos),dataList(test,neg,TestNeg),append(TestPos,TestNeg,AllTest_withLabels), maplist(get_acc_likelihood(Hs_noDuplicate),AllTest_withLabels,PAs,LogLikelihoods), sumList(PAs,PA0),length(AllTest_withLabels,NumTestExs),PA is 100*PA0/NumTestExs, sumList(LogLikelihoods,LogLikelihood), length(Hs_noDuplicate,N),AveTimeTaken is TimeTaken/(N*1000), portray_clause(numSampledHs(N)), tell('onePA.pl'), write(PA),write(' '), told, tell('oneLogLikehood.pl'), write(LogLikelihood),write(' '), told, tell('oneTime.pl'), write(AveTimeTaken),write(' '), told, write(result_oneSample(Problem_info,SampleSize,SameSizeRepeated,PA,LogLikelihood,AveTimeTaken)),writeln('.'),nl,!. get_acc_likelihood(Hs,Ex-TargetLabel,PA,LogLikelihood):- bayes_predict(Ex,Hs,[],Prediction),portray_clause(prediction(Ex-TargetLabel,Prediction)), predictive_accuracy0(TargetLabel,Prediction,PA), logLikelihood0(TargetLabel,Prediction,LogLikelihood). predictive_accuracy0(TargetLabel,Prediction,Accuracy):- TargetLabel=1,!, (Prediction>0.5 -> Accuracy=1; Accuracy=0 ). predictive_accuracy0(TargetLabel,Prediction,Accuracy):- (Prediction=<0.5 -> %%---*Prediction<0.5 -> Accuracy=1; Accuracy=0 ). logLikelihood0(TargetLabel,Likelihood0,LogLikelihood):- %write(Likelihood0), (TargetLabel=1 -> Likelihood=Likelihood0; Likelihood is abs(1-Likelihood0) % bug in ProbLog to have negative value -- could be due to number overflow ), (Likelihood==0.0-> Likelihood2=0.00000000001; Likelihood2=Likelihood ), LogLikelihood is -log(Likelihood2). unCoverRecord(Hs,[],[]). unCoverRecord(Hs,[Ex|CheckList],NUnCoverList):- (set(prediction,bayes_predict)-> bayes_predict(Ex,Hs,[],Prob); slp_predict(Ex,Hs,Prob) ), portray_clause(predicted_result(Ex,Prob)), (Prob>0.5-> NUnCoverList=UnCoverList; NUnCoverList=[Ex|UnCoverList] ), unCoverRecord(Hs,CheckList,UnCoverList). tag_label(Label,Ex,Ex-Label). dataList(test,pos,TestPos_withLabel):- set(test_size,TestSize),TestPosSize is TestSize//2, set(train_size,TrainSize),TrainPosSize is TrainSize//2, %write('%%%%%dataList'),write({TrainPosSize,TestPosSize}),nl, findall(Seq,ex(Seq,1),AllPos), StartIndex is TrainPosSize+1,EndIndex is StartIndex+TestPosSize-1, %write('%%%%%dataList%'),write({StartIndex,EndIndex}),nl, numbersList(StartIndex,EndIndex,TestPosIndexes), fetchbyIndex(AllPos,TestPosIndexes,TestPos), maplist(tag_label(1),TestPos,TestPos_withLabel). dataList(test,neg,TestNeg_withLabel):- set(test_size,TestSize),TestNegSize is TestSize//2, set(train_size,TrainSize),TrainNegSize is TrainSize//2, findall(Seq,ex(Seq,0),AllNeg), StartIndex is TrainNegSize+1,EndIndex is StartIndex+TestNegSize-1, % write('%'),write({StartIndex,EndIndex}), numbersList(StartIndex,EndIndex,TestNegIndexes), fetchbyIndex(AllNeg,TestNegIndexes,TestNeg), maplist(tag_label(0),TestNeg,TestNeg_withLabel). oneSample(SampleSize,SameSizeRepeated,TrainPos,TrainNeg):- set(sampleSizes,Sizes),last_ele(Sizes,Rest,MaxSize),Distance is MaxSize//2, % write('##########'),write(Sizes),write(','),write(MaxSize),write(','),write(Distance),nl, StartIndex is ((SameSizeRepeated-1)*Distance+1), EndIndex is (StartIndex+(SampleSize//2-1)), % write('##########'),write(SameSizeRepeated),write(','),write(StartIndex),write(','),write(EndIndex),nl, numbersList(StartIndex,EndIndex,Indexes), findall(Seq,ex(Seq,1),AllPos), fetchbyIndex(AllPos,Indexes,TrainPos), findall(Seq,ex(Seq,0),AllNeg), fetchbyIndex(AllNeg,Indexes,TrainNeg). oneSampleF(SampleSize,SameSizeRepeated,TrainPos,TrainNeg):- %Size>=4 write('%%%%%oneSampleF'),write({SampleSize,SameSizeRepeated}),nl, %trace, set(sampleSizes,Sizes),last_ele(Sizes,Rest,MaxSize),Distance is MaxSize//2, % write('##########oneSampleF'),write(Sizes),write(','),write(MaxSize),write(','),write(Distance),nl, StartIndex is ((SameSizeRepeated-1)*Distance+1), EndIndex is (StartIndex+(SampleSize//2-1)), % write('%%%%%%%%%oneSampleF'),write(SameSizeRepeated),write(','),write(StartIndex),write(','),write(EndIndex),nl, numbersList(StartIndex,EndIndex,Indexes), %write('##########'),writeln(Indexes), findall(Seq,ex(Seq,1),AllPos), fetchbyIndex(AllPos,Indexes,TrainPos), findall(Seq,ex(Seq,0),AllNeg), fetchbyIndex(AllNeg,Indexes,TrainNeg). %write('********'),write(AllNeg),write(Indexes),writeln(TrainNeg). group_bySize([],[]). group_bySize([hyp(Prob,Hyp)|Hs],GroupedHs):- group_bySize(Hs,PreGroupedHs), length(Hyp,Size), (append(Pre,[Size-SameSizeHs|Post],PreGroupedHs)-> append(Pre,[Size-[hyp(1,Hyp)|SameSizeHs]|Post],GroupedHs); GroupedHs=[Size-[hyp(1,Hyp)]|PreGroupedHs] ). renormalise_blocks(GroupedHs,RenormalisedGroupedHs):- maplist(getBlockPrior,GroupedHs,BlockPriors,PrioredGroupedHs), sumList(BlockPriors,RenormalisedConstant), maplist(renormalise_blocks0(RenormalisedConstant),PrioredGroupedHs,RenormalisedGroupedHs). /* knownBlockPrior(3,1/200):-!. knownBlockPrior(4,4/200):-!. knownBlockPrior(5,14/200):-!. knownBlockPrior(6,27/200):-!. knownBlockPrior(7,50/200):-!. knownBlockPrior(8,47/200):-!. knownBlockPrior(9,47/200):-!. knownBlockPrior(10,10/200):-!. knownBlockPrior(X,0). */ knownBlockPrior(HSize,BlockPrior):- %BlockPrior is 1/(HSize*HSize). BlockPrior is 1/(2^HSize). % Exponential %knownBlockPrior(HSize,1). getBlockPrior(HSize-Hs,BlockPrior,BlockPrior-Hs):- knownBlockPrior(HSize,BlockPrior). renormalise_blocks0(RenormalisedConstant,BlockPrior0-Hs,BlockPrior-Hs):- BlockPrior is BlockPrior0/RenormalisedConstant. bayes_blockPredict(X,BK,BlockPrioredHs,Prob):- maplist(bayes_blockPredict0(X-BK),BlockPrioredHs,WeightedPrediction), sumList(WeightedPrediction,Prob). %the weigth need to renormalised, otherwise bayes_blockPredict0(X-BK,BlockPrior-Hs0,BlockPrior*Prob):- remove_duplicates(Hs0,Hs), bayes_predict(X,Hs,BK,Prob). unCoverRecord_block(Hs,[],[]). unCoverRecord_block(Hs,[Ex|CheckList],NUnCoverList):- bk(0,BK), bayes_blockPredict(Ex,BK,Hs,Prob),portray_clause(predicted_result(Ex,Prob)), (Prob>0.5-> NUnCoverList=UnCoverList; NUnCoverList=[Ex|UnCoverList] ), unCoverRecord_block(Hs,CheckList,UnCoverList).