innng/gremlin

[2018] GRammar-based Evolutionary Machine LearnINg

Epochx-Weka

Anotações

• Os parâmetros de um classificador selecionado por outro são antecedidos por --
• Os parâmetros e o nome de um algoritmo que não é classificador vem dentro de " "
• Parece que uma variável chamada "validProductions" é negativa em certo momento
• Warning no valor alto do pruning, por quê?

Melhorias que podem ser adicionadas:

• Espécies
• Fitness Sharing
• Co-evolução (cooperativa ou competitiva?)
• Opção de selecionar dataset pra treino e teste

Algoritmos e parâmetros

• Parâmetros

  • Geral
    • batchSize (-batch-size)
    • debug (-output-debug-info)
    • doNotCheckCapabilities (-do-not-check-capabilities)
    • numDecimalPlaces (-num-decimal-places)
    • outputOutOfBagComplexityStatistics (-output-out-of-bag-complexity-statistics)
    • numExecutionSlots (-num-slots)
    • printClassifiers (-print)
    • representCopiesUsingWeights (-represent-copies-using-weights)
    • storeOutOfBagPredictions (-store-out-of-bag-predictions)
    • doNotPrintModels (-do-not-print)
    • computeAttributeImportance (-attribute-importance)
    • doNotMakeSplitPointActualValue (-doNotMakeSplitPointActualValue)

• Algoritmos

  • Attribute Selection
    • Eval
      • CfsSubsetEval
        • locallyPredictive (-L)
          • (true, false; false)
        • missingSeparate (-M)
          • (true, false; false)
        • numThreads (-E)
        • poolSize (-P)
        • preComputeCorrelationMatrix (-Z)
    • Search
      • BestFirst
        • direction (-D)
          • (0, 1, 2; 1)
        • lookupCacheSize (-?)
        • searchTerminatior (-N)
          • ([2, 10]; 5) inteiro
        • startSet (-P)
      • GreedyStepwise
        • conservativeForwardSelection (-C)
          • (true, false; false)
        • debuggingOutput (-D)
        • generateRanking (-R)
          • (true, false; false)
        • numToSelect (-N)
          • ([10, 1000]; 30) inteiro
        • searchBackwards (-B)
          • (true, false; false)
        • startSet (-P)
        • threshold (-T)
        • regras:
          • N se R false
  • Base
    • BayesNet
      • estimator (-E)
      • searchAlgorithm (-Q)
        • (K2, HillClimber, LAGDHillClimber, SimulatedAnnealing, TabuSearch, TAN; K2)
      • useADTree (-D)
        • (true, false; false)
    • NaiveBayes
      • displayModelInOldFormat (-O)
      • useKernelEstimator (-K)
        • (true, false; false)
      • useSupervisedDiscretization (-D)
        • (true, false; false)
      • regras:
        • K se D false
    • NaiveBayesMultinomial
    • Logistic
      • maxIts (-M)
      • ridge (-R)
        • ([1e-12, 10]; 1e-7)
      • useConjugateGradientDescent (-C)
    • MultilayerPerceptron
      • GUI (-G)
      • autoBuild (-?)
      • decay (-D)
        • (true, false; false)
      • hiddenLayers (-H)
        • (a, i, o, t; a)
      • learningRate (-L)
        • ([0.1, 1]; 0.3)
      • momentum (-M)
        • ([0.1, 1]; 0.2)
      • nominalToBinaryFilter (-B)
        • (true, false; false)
      • normalizeAttributes (-I)
      • normalizeNumericClass (-C)
        • (true, false; false)
      • reset (-R)
        • (true, false; false)
      • seed (-S)
        • (1; 1)
      • trainingTime (-N)
      • validationSetSize (-V)
      • validationThreshold (-E)
    • SGD
      • dontNormalize (-N)
        • (true, false; false)
      • dontReplaceMissing (-M)
        • (true, false; false)
      • epochs (-E)
      • epsilon (-C)
      • lambda (-R)
        • ([1e-12, 10]; 1e-4)
      • learningRate (-L)
        • ([0.00001, 0.1]; 0.01)
      • lossFunction (-F)
        • (0, 1, 2; 0)
      • seed (-S)
    • SimpleLogistic
      • errorOnProbabilities (-P)
      • heuristicStop (-H)
      • maxBoostingIterations (-M)
      • numBoostingIterations (-I)
      • useAIC (-A)
        • (true, false; false)
      • useCrossValidation (-S)
        • (true, false; false)
      • weightTrimBeta (-W)
        • W_HIDDEN (0, 1; 0)
          • 1_W (0; 0)
          • 2_W ([0, 1]; 0)
      • regras:
        • 1_W se W_HIDDEN 0
        • 2_W se W_HIDDEN 1
    • SMO
      • buildCalibrationModels (-M)
      • c (-C)
        • ([0.5, 1.5]; 1)
      • calibrator (-calibrator)
      • checksTurnedOff (-no-checks)
      • epsilon (-P)
      • filterType (-N)
        • (0, 1, 2; 0)
      • kernel (-K)
        • (NormalizedPolyKernel, PolyKernel, Puk, RBFKernel; NormalizedPolyKernel)
        • NormalizedPolyKernel
          • exponent (-E)
            • ([0.2, 5]; 1)
          • useLowerOrder (-L)
            • (true, false; false)
        • PolyKernel
          • exponent (-E)
            • ([0.2, 5]; 1)
          • useLowerOrder (-L)
            • (true, false; false)
        • Puk
          • omega (-O)
            • ([0.1, 1]; 1)
          • sigma (-S)
            • ([0.1, 10]; 1)
        • RBFKernel
          • gamma (-G)
            • ([0.0001, 1]; 0.01)
      • numFolds (-V)
      • randomSeed (-W)
      • toleranceParameter (-L)
      • regras:
        • parâmetros próprios de cada algoritmo dependendo do valor de K
    • VotedPerceptron
      • exponent (-E)
        • ([0.2, 5]; 1)
      • maxK (-M)
        • ([5000, 50000]; 10000) inteiro
      • numIterations (-I)
        • ([1, 10]; 1) inteiro
      • seed (-S)
    • IBk
      • KNN (-K)
        • ([1, 64]; 1) inteiro
      • crossValidate (-X)
        • (true, false; false)
      • distanceWeighting (-I/-F)
        • (true, false; false)
      • meanSquare (-E)
        • (true, false; false)
      • nearestNeightbourSearchAlgorithm (-A)
      • windowSize (-W)
      • regras:
        • F se I false
    • KStar
      • entropicAutoBlend (-E)
        • (true, false; false)
      • globalBlend (-B)
        • ([1, 100]; 20) inteiro
      • missingMode (-M)
        • (a, d, m, n; a)
    • DecisionTable
      • crossVal (-X)
        • (1, 2, 3, 4; 1)
      • evaluationMeasure (-E)
        • (acc, rmse, mae, aux; acc)
      • search (-S)
        • (BestFirst, GreedyStepwise; BestFirst)
      • useIBk (-I)
        • (true, false; false)
    • JRip
      • checkErrorRate (-E)
        • (true, false; false)
      • folds (-F)
      • minNo (-N)
        • ([1, 5]; 2)
      • optimizations (-O)
        • ([1, 5]; 2) inteiro
      • seed (-S)
      • usePruning (-P)
        • (true, false; false)
    • OneR
      • minBucketSize (-B)
        • ([1, 32]; 6) inteiro
    • PART
      • binarySplits (-B)
        • (true, false; false)
      • confidenceFactor (-C)
      • minNumObj (-M)
        • ([1, 64]; 2) inteiro
      • numFolds (-N)
        • ([2, 5]; 3) inteiro
      • reducedErrorPruning (-R)
        • (true, false; false)
      • seed (-Q)
      • unpruned (-U)
      • useMDLcorrection (-J)
      • regras:
        • N se R true
    • ZeroR
    • DecisionStump
    • J48
      • binarySplits (-B)
        • (true, false; false)
      • collapseTree (-O)
        • (true, false; false)
      • confidenceFactor (-C)
        • C_HIDDEN ([0, 1]; 0.25)
          • C ([0, 1]; 0.25)
      • minNumObj (-M)
        • ([1, 64]; 2) inteiro
      • numFolds (-N)
      • reducedErrorPruning (-R)
      • saveInstanceData (-L)
      • seed (-?)
      • subtreeRaising (-S)
        • (true, false; false)
      • unpruned (-U)
        • (true, false; false)
      • useLaplace (-A)
        • (true, false; false)
      • useMDLcorrection (-J)
        • (true, false; false)
      • regras:
        • U se S false
        • U se C_HIDDEN 0
        • C se C_HIDDEN 1
    • LMT
      • convertNominal (-B)
        • (true, false; false)
      • errorOnProbabilities (-P)
        • (true, false; false)
      • fastRegression (-C)
        • (true, false; false)
      • minNumInstances (-M)
        • ([1, 64]; 15) inteiro
      • numBoostingIterations (-I)
      • splitOnResiduals (-R)
        • (true, false; false)
      • useAIC (-A)
        • (true, false; false)
      • weightTrimBeta (-W)
        • W_HIDDEN (0, 1; 0)
          • 1_W (0; 0)
          • 2_W ([0, 1]; 0)
      • regras:
        • 1_W se W_HIDDEN 0
        • 2_W se W_HIDDEN 1
    • RandomForest
      • bagSizePercent (-P)
      • breakTiesRandomly (-B)
      • calcOutOfBag (-O)
      • maxDepth (-depth)
        • depth_HIDDEN (0, 1; 0)
          • 1_INT_depth (0; 0)
          • 2_INT_depth ([1, 20]; 2) inteiro
      • numFeatures (-K)
        • features_HIDDEN (0, 1; 0)
          • 1_INT_K(0; 0)
          • 2_INT_K ([1, 32]; 2) inteiro
      • numIterations (-I)
        • ([2, 256]; 10)
      • seed (-S)
      • regras:
        • 1_INT_K se features_HIDDEN 0
        • 2_INT_K se features_HIDDEN 1
        • 1_INT_depth se depth_HIDDEN 0
        • 2_INT_depth se depth_HIDDEN 1
    • RandomTree
      • KValue (-K)
        • features_HIDDEN (0, 1; 0)
          • 1_INT_K (0; 0)
          • 2_INT_K ([2, 32]; 2) inteiro
      • allowUnclassifiedInstances (-U)
        • (true, false; false)
      • breakTiesRandomly (-B)
      • maxDepth (-depth)
        • depth_HIDDEN (0, 1; 0)
          • 1_INT_depth (0; 0)
          • 2_INT_depth ([2, 20]; 2) inteiro
      • minNum (-M)
        • ([1, 64]; 1) inteiro
      • minVarianceDrop (-V)
      • numFolds (-N)
        • back_HIDDEN (0, 1; 0)
          • 1_INT_N (0; 0)
          • 2_INT_N ([2, 5]; 3) inteiro
      • seed (-S)
      • regras:
        • 1_INT_K se features_HIDDEN 0
        • 2_INT_K se features_HIDDEN 1
        • 1_INT_depth se depth_HIDDEN 0
        • 2_INT_depth se depth_HIDDEN 1
        • 1_INT_N se back_HIDDEN 0
        • 2_INT_N se back_HIDDEN 1
    • REPTree
      • initialCount (-I)
      • maxDepth (-L)
        • depth_HIDDEN (0, 1; 0)
          • 1_INT_L (-1; -1)
          • 2_INT_L ([2, 20]; 2)
      • minNum (-M)
        • ([1, 64]; 2) inteiro
      • minVarianceDrop (-V)
        • ([1e-5, 1e-1]; 1e-3)
      • noPruning (-P)
        • (true, false; false)
      • numFolds (-N)
      • seed (-S)
      • spreadInitialCount (-R)
      • regras:
        • 1_INT_L se depth_HIDDEN 0
        • 2_INT_L se depth_HIDDEN 1
  • Ensemble
    • Stacking
      • classifiers (-B)
      • metaClassifier (-M)
      • numFolds (-X)
        • (10; 10)
      • seed (-S)
        • (1; 1)
    • Vote
      • classifiers (-B)
      • combinationRule (-R)
        • (AVG, PROD, MAJ, MIN, MAX; AVG)
      • preBuiltClassifiers (-P)
      • seed (-S)
        • (1; 1)
  • Meta
    • LWL
      • KNN (-K)
        • K_HIDDEN (0, 1; 0)
          • 1_K (-1, 10, 30, 60, 90, 120; -1)
      • classifier (-W)
      • nearestNeighbourSearchAlgorithm (-A)
        • (LinearNNSearch; LinearNNSearch)
      • weightingKernel (-U)
        • U_HIDDEN (0, 1; 0)
          • 1_U (0, 1, 2, 3, 4; 0)
      • regras:
        • 1_K se K_HIDDEN 1
        • 1_U se K_HIDDEN 0
        • 1_U se U_HIDDEN 1
        • 1_K se U_HIDDEN 0
    • AdaBoostM1
      • classifier (-W)
      • numIterations (-I)
        • ([2, 128]; 10) inteiro
      • seed (-S)
        • (1; 1)
      • useResampling (-Q)
        • (true, false; false)
      • weightThreshold (-P)
        • p_HIDDEN (0, 1; 0)
          • 1_P (100; 100)
          • 2_INT_P ([50, 100]; 100)
        • regras:
          • 1_P se p_HIDDEN 0
          • 2_INT_P se p_HIDDEN 1
    • AttributeSelectedClassifier
      • classifier (-W)
      • evaluator (-E)
        • (CfsSubsetEval; CfsSubsetEval)
      • search (-S)
        • (BestFirst, GreedyStepwise; BestFirst)
    • Bagging
      • bagSizePercent (-P)
        • ([10, 100]; 100) inteiro
      • calcOutOfBag (-O)
        • (true, false; false)
      • classifier (-W)
      • numIterations (-I)
        • ([2, 128]; 10) inteiro
      • seed (-1)
        • (1; 1)
      • regras:
        • O se INT_P 100
    • RandomComittee
      • classifier (-W)
      • numIterations (-I)
        • ([2, 64]; 10) inteiro
      • seed (-S)
        • (1; 1)
    • RandomSubSpace
      • classifier (-W)
      • numIterations (-I)
        • ([2, 64]; 10) inteiro
      • seed (-S)
        • (1; 1)
      • subSpaceSize (-P)
        • ([0.1, 1]; 0.5)