|
// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.
using System.Text;
using Microsoft.ML.CommandLine;
using Microsoft.ML.Runtime;
using Microsoft.ML.TestFramework;
using Microsoft.ML.Trainers.FastTree;
namespace Microsoft.ML.RunTests
{
//=========================== Binary classifiers ====================
public class PredictorAndArgs
{
public SubComponent Trainer { get; set; }
public SubComponent Scorer { get; set; }
public SubComponent Tester { get; set; }
public string Tag { get; set; }
public string[] MamlArgs;
// Whether to write the progress to the output console file or not.
public bool BaselineProgress { get; set; }
public PredictorAndArgs()
{
}
public PredictorAndArgs(SubComponent trainer, string tag = null)
{
Trainer = trainer;
Tag = tag;
}
}
public class TestLearnersBase
{
// This ensures that the needed assemblies are loaded!
static TestLearnersBase()
{
bool ok = true;
ok &= typeof(FastTreeBinaryTrainer) != null;
Contracts.Check(ok, "Missing assemblies!");
}
// New.
public static PredictorAndArgs binaryPrior = new PredictorAndArgs
{
Trainer = new SubComponent("PriorPredictor"),
Tag = "BinaryPrior"
};
// New.
public static PredictorAndArgs binarySdca = new PredictorAndArgs
{
Trainer = new SubComponent("SDCA", "maxIterations=5 checkFreq=9 nt=1"),
Tag = "BinarySDCA"
};
// New.
public static PredictorAndArgs binarySdcaL1 = new PredictorAndArgs
{
Trainer = new SubComponent("SDCA", "l2=1e-06 l1=0.5 maxIterations=5 checkFreq=9 nt=1"),
Tag = "BinarySDCA-L1"
};
// New.
public static PredictorAndArgs binarySdcaSmoothedHinge = new PredictorAndArgs
{
Trainer = new SubComponent("SDCA", "l2=1e-06 loss=SmoothedHinge l1=0.5 maxIterations=5 checkFreq=9 nt=1"),
Tag = "BinarySDCA-SmoothedHinge"
};
// New.
public static PredictorAndArgs binarySgd = new PredictorAndArgs
{
Trainer = new SubComponent("SGD", "maxIterations=2 checkFreq=9 nt=1"),
Tag = "BinarySGD"
};
// New.
public static PredictorAndArgs binarySgdHinge = new PredictorAndArgs
{
Trainer = new SubComponent("SGD", "loss=Hinge maxIterations=2 checkFreq=9 nt=1"),
Tag = "BinarySGD-Hinge"
};
// New.
public static PredictorAndArgs multiclassSdca = new PredictorAndArgs
{
Trainer = new SubComponent("SDCAMC", "maxIterations=20 checkFreq=0 nt=1"),
Tag = "MultiClassSDCA"
};
// New.
public static PredictorAndArgs multiclassSdcaL1 = new PredictorAndArgs
{
Trainer = new SubComponent("SDCAMC", "l2=1e-04 l1=0.5 maxIterations=20 checkFreq=0 nt=1"),
Tag = "MulticlassSDCA-L1"
};
// New.
public static PredictorAndArgs multiclassSdcaSmoothedHinge = new PredictorAndArgs
{
Trainer = new SubComponent("SDCAMC", "l2=1e-04 loss=SmoothedHinge maxIterations=20 checkFreq=0 nt=1"),
Tag = "MulticlassSDCA-SmoothedHinge"
};
// Old.
public static PredictorAndArgs perceptron_tlOld = new PredictorAndArgs
{
Trainer = new SubComponent("AveragedPerceptron", "lr=0.01 iter=100 numnorm=500 lazy+"),
};
public static PredictorAndArgs perceptron = new PredictorAndArgs
{
Trainer = new SubComponent("AveragedPerceptron", "lr=0.01 iter=100 lazy+"),
};
// New.
public static PredictorAndArgs perceptron_reg = new PredictorAndArgs
{
Trainer = new SubComponent("AveragedPerceptron", "lr=0.01 iter=100 lazy+ reg=0.002"),
Tag = "AveragedPerceptron-Reg"
};
public static PredictorAndArgs Ova = new PredictorAndArgs
{
Trainer = new SubComponent("OVA", "p=AvgPer{ lr=0.8 }"),
MamlArgs = new[] { "norm=no" },
};
public static PredictorAndArgs OvaWithFastForest = new PredictorAndArgs
{
Trainer = new SubComponent("OVA", "p=FastForest{ }"),
MamlArgs = new[] { "norm=no" },
Tag = "OVA-FastForest",
};
public static PredictorAndArgs Pkpd = new PredictorAndArgs
{
Trainer = new SubComponent("PKPD", "p=AvgPer { lr=0.8 }"),
MamlArgs = new[] { "norm=no" },
};
// Old.
public static PredictorAndArgs perceptronDefault = new PredictorAndArgs
{
Trainer = new SubComponent("AveragedPerceptron"),
};
// Old.
public static PredictorAndArgs perceptronNotNorm = new PredictorAndArgs
{
Trainer = new SubComponent("AveragedPerceptron", "norm={}"),
};
// New and Old.
public static PredictorAndArgs linearSVM = new PredictorAndArgs
{
Trainer = new SubComponent("LinearSVM", "iter=100 lambda=0.03"),
};
// Old.
public static PredictorAndArgs linearSVMNotNorm = new PredictorAndArgs
{
Trainer = new SubComponent("LinearSVM", "iter=100 lambda=0.03 norm={}"),
};
// Old.
public static PredictorAndArgs linearSVMNotNormOne = new PredictorAndArgs
{
Trainer = new SubComponent("LinearSVM", "iter=1 lambda=0.03 norm={}"),
};
// Old.
public static PredictorAndArgs logisticRegression_tlOld = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-3 norm={} nt=1"),
};
// New.
public static PredictorAndArgs logisticRegression = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-3 nt=1"),
MamlArgs = new[] { "norm=no" },
BaselineProgress = true
};
// New.
public static PredictorAndArgs symSGD = new PredictorAndArgs
{
Trainer = new SubComponent("SymSGD", "nt=1"),
MamlArgs = new[] { "norm=no" },
BaselineProgress = true
};
// New.
public static PredictorAndArgs logisticRegressionNonNegative = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-4 nt=1 nn=+"),
MamlArgs = new[] { "norm=no" },
Tag = "LogisticRegression-non-negative",
BaselineProgress = true
};
// New.
public static PredictorAndArgs logisticRegressionNorm = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-3 nt=1"),
Tag = "LogisticRegression-norm",
BaselineProgress = true
};
// New.
public static PredictorAndArgs logisticRegressionBinNorm = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-3 nt=1"),
MamlArgs = new[] { "xf=BinNormalizer{col=Features numBins=5}" },
Tag = "LogisticRegression-bin-norm",
BaselineProgress = true
};
// Old.
public static PredictorAndArgs logisticRegressionSGD = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-3 sgd=0.1 nt=1"),
Tag = "SGD",
};
// New.
public static PredictorAndArgs logisticRegressionGaussianNorm = new PredictorAndArgs
{
Trainer = new SubComponent("LogisticRegression", "l1=1.0 l2=0.1 ot=1e-3 nt=1"),
MamlArgs = new[] { "xf=MeanVarNormalizer{col=Features}" },
Tag = "LogisticRegression-GaussianNorm",
BaselineProgress = true
};
// New.
public static PredictorAndArgs multiclassLogisticRegression = new PredictorAndArgs
{
Trainer = new SubComponent("MulticlassLogisticRegression", "l1=0.001 l2=0.1 ot=1e-3 nt=1"),
MamlArgs = new[] { "norm=no" },
BaselineProgress = true,
};
// New.
public static PredictorAndArgs multiclassLogisticRegressionNonNegative = new PredictorAndArgs
{
Trainer = new SubComponent("MulticlassLogisticRegression", "l1=0.001 l2=0.1 ot=1e-3 nt=1 nn=+"),
MamlArgs = new[] { "norm=no" },
Tag = "LogisticRegression-Non-Negative",
BaselineProgress = true,
};
// Old.
public static PredictorAndArgs multiclassLogisticRegressionRegularized = new PredictorAndArgs
{
Trainer = new SubComponent("MulticlassLogisticRegression", "nt=1"),
BaselineProgress = true,
};
// New.
public static PredictorAndArgs FastForestClassification = new PredictorAndArgs
{
Trainer = new SubComponent("FastForestClassification", "nl=5 mil=10 iter=10"),
BaselineProgress = true,
};
// New and Old.
public static PredictorAndArgs fastRankClassification = new PredictorAndArgs
{
Trainer = new SubComponent("FastRank", "nl=5 mil=5 lr=0.25 iter=20"),
BaselineProgress = true,
};
// New.
public static PredictorAndArgs fastRankClassificationPruning = new PredictorAndArgs
{
Trainer = new SubComponent("FastRank", "nl=5 mil=5 lr=0.25 iter=20 pruning=+"),
Tag = "FastRank-Pruning",
BaselineProgress = true,
};
public static PredictorAndArgs fastRankClassificationWeighted = new PredictorAndArgs
{
Trainer = new SubComponent("FastRank", "nl=5 mil=20 lr=0.25 iter=20"),
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeClassfier = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "nl=5 mil=5 lr=0.25 iter=20 mb=255"),
Tag = "FastTree",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMClassifier = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBMBinary", "nt=1 nl=5 mil=5 lr=0.25 iter=20 mb=255"),
Tag = "LightGBM",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMGoss = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBM", "nt=1 iter=10 v=+ booster=goss lr=0.2 mil=10 nl=20"),
Tag = "LightGBMGoss",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMDart = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBM", "nt=1 iter=10 booster=dart lr=0.2 mil=10 nl=20"),
Tag = "LightGBMDart",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMMC = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBMMC", "nt=1 iter=10 v=- lr=0.2 mil=10 nl=20"),
Tag = "LightGBMMC",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMReg = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBMR", "nt=1 iter=50 v=+ booster=gbdt{l1=0.2 l2=0.2} lr=0.2 mil=10 nl=20"),
Tag = "LightGBMReg",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMRegMae = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBMR", "nt=1 iter=50 em=MeanAbsoluteError v=+ lr=0.2 mil=10 nl=20"),
Tag = "LightGBMRegMae",
BaselineProgress = true,
};
public static PredictorAndArgs LightGBMRegRmse = new PredictorAndArgs
{
Trainer = new SubComponent("LightGBMR", "nt=1 iter=50 em=RootMeanSquaredError v=+ lr=0.2 mil=10 nl=20"),
Tag = "LightGBMRegRmse",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeWithCategoricalClassfier = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "cat=+ nl=5 mil=5 lr=0.25 iter=20 mb=255"),
Tag = "FastTreeCategorical",
BaselineProgress = true
};
public static PredictorAndArgs FastTreeWithCategoricalClassfierDisk = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "cat=+ nl=5 mil=5 lr=0.25 iter=20 mb=255 dt+"),
Tag = "FastTreeCategoricalDisk",
BaselineProgress = true
};
public static PredictorAndArgs FastTreeClassfierHighMinDocs = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "mil=10000 iter=5"),
Tag = "FastTreeHighMinDocs",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeClassfierDisk = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "nl=5 mil=5 lr=0.25 iter=20 mb=255 dt+"),
Tag = "FastTreeDisk",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeDropoutClassfier = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "nl=5 mil=5 tdrop=0.5 lr=0.25 iter=20 mb=255"),
Tag = "FastTreeDrop",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeBsrClassfier = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeBinaryClassification", "nl=5 mil=5 bsr+ lr=0.25 iter=20 mb=255"),
Tag = "FastTreeBsr",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeRanker = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeRanking", "iter=10 nl=10 mil=32 lr=0.30 mb=255"),
Tag = "FastTree",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeRankerCustomGains = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeRanking", "iter=10 nl=10 mil=32 lr=0.30 mb=255 gains=0,2,3,5,6"),
MamlArgs = new[] { "eval=Ranking{gains=0,2,3,5,6}" },
Tag = "FastTreeCustomGains",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeRegressor = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeRegression", "nl=5 mil=5 lr=0.25 iter=20 mb=255"),
Tag = "FastTree",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeRegressorCategorical = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeRegression", "cat=+ nl=5 mil=5 lr=0.25 iter=20 mb=255"),
Tag = "FastTreeCategorical",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeUnderbuiltRegressor = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeRegression", "nl=30 mil=30 lr=0.25 iter=10 mb=255"),
Tag = "FastTree",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeDropoutRegressor = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeRegression", "nl=5 mil=5 tdrop=0.5 lr=0.25 iter=20 mb=255"),
Tag = "FastTreeDrop",
BaselineProgress = true,
};
public static PredictorAndArgs FastTreeTweedieRegressor = new PredictorAndArgs
{
Trainer = new SubComponent("FastTreeTweedieRegression", "nl=5 mil=5 lr=0.25 iter=10 mb=255 index=1.5"),
Tag = "FastTreeTweedie",
MamlArgs = new[] { "Eval=regression{loss=tweedie{index=1.5}}" },
BaselineProgress = true,
};
public static PredictorAndArgs RegressionGamTrainer = new PredictorAndArgs
{
Trainer = new SubComponent("RegressionGamTrainer", ""),
Tag = "RegressionGamTrainer",
BaselineProgress = true,
};
public static PredictorAndArgs BinaryClassificationGamTrainer = new PredictorAndArgs
{
Trainer = new SubComponent("BinaryClassificationGamTrainer", ""),
Tag = "BinaryClassificationGamTrainer",
BaselineProgress = true,
};
public static PredictorAndArgs BinaryClassificationGamTrainerDiskTranspose = new PredictorAndArgs
{
Trainer = new SubComponent("BinaryClassificationGamTrainer", "dt+"),
Tag = "BinaryClassificationGamTrainerDiskTranspose",
BaselineProgress = true,
};
// New.
public static PredictorAndArgs QuantileRegressionScorer = new PredictorAndArgs
{
Trainer = new SubComponent("FastForestRegression", "nl=5 mil=5 iter=20"),
Scorer = new SubComponent("QuantileRegression", "quantiles = 0.25,0.5,0.75"),
Tag = "QuantileRegressorTester",
BaselineProgress = true,
};
// New.
public static PredictorAndArgs FastForestRegression = new PredictorAndArgs
{
Trainer = new SubComponent("FastForestRegression", "nl=5 mil=5 iter=20"),
BaselineProgress = true,
};
public static PredictorAndArgs fastRankRegression = new PredictorAndArgs
{
Trainer = new SubComponent("FastRankRegression", "nl=5 mil=5 lr=0.25 iter=20"),
BaselineProgress = true,
};
public static PredictorAndArgs fastRankRegressionWeighted = new PredictorAndArgs
{
Trainer = new SubComponent("FastRankRegression", "nl=5 mil=20 lr=0.25 iter=20"),
BaselineProgress = true,
};
public static PredictorAndArgs OGD = new PredictorAndArgs
{
Trainer = new SubComponent("OGD", "iter=100"),
MamlArgs = new[] { "xf=MinMax{col=Features zero-}" }
};
public static PredictorAndArgs Ols = new PredictorAndArgs
{
Trainer = new SubComponent("OLS"),
MamlArgs = new[] { "norm=no" },
};
public static PredictorAndArgs OlsNorm = new PredictorAndArgs
{
Trainer = new SubComponent("OLS"),
MamlArgs = new[] { "xf=MinMax{col=Features}" },
Tag = "OLSNorm"
};
public static PredictorAndArgs OlsReg = new PredictorAndArgs
{
Trainer = new SubComponent("OLS", "l2=0.1"),
MamlArgs = new[] { "norm=no" },
Tag = "OLSReg"
};
public static PredictorAndArgs Sdcar = new PredictorAndArgs
{
Trainer = new SubComponent("SDCAR", "nt=1"),
Tag = "SDCAR"
};
public static PredictorAndArgs SdcarNorm = new PredictorAndArgs
{
Trainer = new SubComponent("SDCAR", "l2=0.01 l1=0 iter=10 checkFreq=-1 nt=1"),
Tag = "SDCARNorm"
};
public static PredictorAndArgs SdcarReg = new PredictorAndArgs
{
Trainer = new SubComponent("SDCAR", "l2=0.1 l1=0 iter=10 checkFreq=-1 nt=1"),
Tag = "SDCARReg"
};
public static PredictorAndArgs fastRankRanking = new PredictorAndArgs
{
Trainer = new SubComponent("FastRankRanking", "iter=10 nl=10 mil=32 lr=0.30"),
BaselineProgress = true,
};
public static PredictorAndArgs fastRankRankingWeighted = new PredictorAndArgs
{
Trainer = new SubComponent("FastRankRanking", "iter=10 nl=10 mil=32 lr=0.30"),
BaselineProgress = true,
};
public static PredictorAndArgs poissonRegression = new PredictorAndArgs
{
Trainer = new SubComponent("PoissonRegression", "l1=1.0 ot=1e-2 nt=1"),
Tester = new SubComponent("Regression", "loss=PoissonLoss"),
MamlArgs = new[] { "xf=MinMax{col=Features zero-}" },
BaselineProgress = true
};
public static PredictorAndArgs poissonRegressionNonNegative = new PredictorAndArgs
{
Trainer = new SubComponent("PoissonRegression", "l1=1.0 ot=1e-2 nt=1 nn=+"),
Tester = new SubComponent("Regression", "loss=PoissonLoss"),
MamlArgs = new[] { "xf=MinMax{col=Features zero-}" },
Tag = "PoissonRegression-Non-Negative",
BaselineProgress = true
};
public static PredictorAndArgs NnBinDefault = new PredictorAndArgs
{
Trainer = new SubComponent("BinaryNeuralNetwork", "accel=sse"),
Tag = "Default",
};
public static PredictorAndArgs NnBinNoNorm = new PredictorAndArgs
{
Trainer = new SubComponent("BinaryNeuralNetwork", "accel=sse"),
MamlArgs = new[] { "norm=no" },
Tag = "NoNorm",
};
public static PredictorAndArgs NnBinMomentum = new PredictorAndArgs
{
Trainer = new SubComponent("BinaryNeuralNetwork", "momentum=0.5 accel=sse"),
Tag = "Momentum",
};
public static PredictorAndArgs NnBinCustom(string path)
{
return new PredictorAndArgs
{
Trainer = new SubComponent("BinaryNeuralNetwork", "filename={" + path + "} lr=0.01 accel=sse"),
Tag = "Custom",
};
}
public static PredictorAndArgs NnBinInline(string netDefinition)
{
var quoted = new StringBuilder();
CmdQuoter.QuoteValue(netDefinition, quoted);
return new PredictorAndArgs
{
Trainer = new SubComponent("BinaryNeuralNetwork", "net=" + quoted + " accel=sse"),
Tag = "Inline",
};
}
public static PredictorAndArgs NnMultiDefault(int numOutputs)
{
return new PredictorAndArgs
{
Trainer = new SubComponent("MultiClassNeuralNetwork", string.Format("lr=0.1 accel=sse output={0}", numOutputs)),
Tag = "Default",
};
}
public static PredictorAndArgs NnMultiCustom(string path)
{
return new PredictorAndArgs
{
Trainer = new SubComponent("MultiClassNeuralNetwork", "filename={" + path + "} lr=0.1 accel=sse"),
Tag = "Default",
};
}
public static PredictorAndArgs NnMultiCustomLred(string path)
{
return new PredictorAndArgs
{
Trainer = new SubComponent("MultiClassNeuralNetwork", "filename={" + path + "} lr=0.1 lred=0.9 lredfreq=5 lrederror=0.02 accel=sse"),
Tag = "Lred",
};
}
public static PredictorAndArgs NnMultiMomentum(int numOutputs, bool useNag = false)
{
return new PredictorAndArgs
{
Trainer = new SubComponent("MultiClassNeuralNetwork",
string.Format("lr=0.1 momentum=0.5 accel=sse output={0} nag{1}", numOutputs, useNag ? "+" : "-")),
Tag = "Momentum" + (useNag ? "Nag" : string.Empty),
};
}
public static PredictorAndArgs DssmDefault(int qryFeaturesCount, int docFeaturesCount, int negativeDocsCount, int numIterations, float gamma)
{
string settings = string.Format("qfeats={0} dfeats={1} negdocs={2} iter={3} gamma={4} accel=sse",
qryFeaturesCount, docFeaturesCount, negativeDocsCount, numIterations, gamma);
return new PredictorAndArgs
{
Trainer = new SubComponent("DssmNet", settings),
Scorer = new SubComponent("MultiClass"),
Tag = "DefaultDssm",
};
}
public static PredictorAndArgs OneClassSvmRbf = new PredictorAndArgs
{
Trainer = new SubComponent("OneClassSVM"),
};
public static PredictorAndArgs OneClassSvmSigmoid = new PredictorAndArgs
{
Trainer = new SubComponent("OneClassSVM", "ker=SigmoidKernel {g=0.005}"),
};
public static PredictorAndArgs OneClassSvmLinear = new PredictorAndArgs
{
Trainer = new SubComponent("OneClassSVM", "ker=LinearKernel"),
};
public static PredictorAndArgs OneClassSvmPoly = new PredictorAndArgs
{
Trainer = new SubComponent("OneClassSVM", "ker=PolynomialKernel {b=1}"),
};
public static PredictorAndArgs PCAAnomalyDefault = new PredictorAndArgs
{
Trainer = new SubComponent("pcaAnomaly"),
Tag = "Default"
};
public static PredictorAndArgs PCAAnomalyNoNorm = new PredictorAndArgs
{
Trainer = new SubComponent("pcaAnomaly"),
MamlArgs = new[] { "norm=no" },
Tag = "NoNorm"
};
public static PredictorAndArgs LDSVMDefault = new PredictorAndArgs
{
Trainer = new SubComponent("LdSvm", "iter=1000"),
Tag = "LDSVM-def"
};
public static PredictorAndArgs LDSVMNoBias = new PredictorAndArgs
{
Trainer = new SubComponent("LdSvm", "iter=1000 bias=-"),
Tag = "LDSVM-nob"
};
public static PredictorAndArgs KMeansDefault = new PredictorAndArgs
{
Trainer = new SubComponent("KM", "nt=1"),
Tag = "KMeans-def"
};
public static PredictorAndArgs KMeansDefaultNoNormalizer = new PredictorAndArgs
{
Trainer = new SubComponent("KM", "norm={} nt=1"),
Tag = "KMeans-def-nonorm"
};
public static PredictorAndArgs KMeansInitPlusPlus = new PredictorAndArgs
{
Trainer = new SubComponent("KM", "nt=1 init=KMeansPlusPlus"),
Tag = "KMeans-initplusplus"
};
public static PredictorAndArgs KMeansInitRandom = new PredictorAndArgs
{
Trainer = new SubComponent("KM", "nt=1 init=Random"),
Tag = "KMeans-initrandom"
};
public static string BinaryTrembleTrainer = "BinaryTremble";
public static string MulticlassTrembleTrainer = "MultiClassTremble";
public static PredictorAndArgs BinaryTrembleDecisionTreeLR = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryTrembleTrainer, "inp=SingleFeaturePredictor inp=LogisticRegression{maxiter=10 quiet=+ t-} lnp=LogisticRegression{maxiter=100 quiet=+ t-} nl=8"),
Tag = "BinaryTremble-lr"
};
public static PredictorAndArgs BinaryDecisionTreeDefault = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryTrembleTrainer, ""),
Tag = "BinaryDT-def"
};
public static PredictorAndArgs BinaryDecisionTreePruning = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryTrembleTrainer, "prune=+"),
Tag = "BinaryDT-prune"
};
public static PredictorAndArgs BinaryDecisionTreeGini = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryTrembleTrainer, "imp=Gini"),
Tag = "BinaryDT-gini"
};
public static PredictorAndArgs BinaryDecisionTreeModified = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryTrembleTrainer, "mil=2 inp=SingleFeaturePredictor{nb=FIFTEEN ff=0.7} nl=40"),
Tag = "BinaryDT-mod"
};
public static PredictorAndArgs BinaryDecisionTreeRewt = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryTrembleTrainer, "mil=2 inp=SingleFeaturePredictor{nb=FIFTEEN ff=0.7} nl=40 rewt=+"),
Tag = "BinaryDT-rewt"
};
public static PredictorAndArgs MulticlassTrembleDecisionTreeLR = new PredictorAndArgs
{
Trainer = new SubComponent(MulticlassTrembleTrainer, "inp=SingleFeaturePredictor inp=LogisticRegression{maxiter=10 quiet=+ t-} lnp=MulticlassLogisticRegression{maxiter=100 quiet=+ t-} nl=8"),
Tag = "MulticlassTremble-lr"
};
public static PredictorAndArgs MulticlassDecisionTreeDefault = new PredictorAndArgs
{
Trainer = new SubComponent(MulticlassTrembleTrainer, "nl=20"),
Tag = "MulticlassDT-def"
};
public static PredictorAndArgs MulticlassDecisionTreePruning = new PredictorAndArgs
{
Trainer = new SubComponent(MulticlassTrembleTrainer, "nl=20 prune=+"),
Tag = "MulticlassDT-prune"
};
public static PredictorAndArgs MulticlassDecisionTreeGini = new PredictorAndArgs
{
Trainer = new SubComponent(MulticlassTrembleTrainer, "nl=20 imp=Gini"),
Tag = "MulticlassDT-gini"
};
public static PredictorAndArgs MulticlassDecisionTreeModified = new PredictorAndArgs
{
Trainer = new SubComponent(MulticlassTrembleTrainer, "mil=2 inp=SingleFeaturePredictor{nb=FIFTEEN ff=0.7} nl=40"),
Tag = "MulticlassDT-mod"
};
public const string BinaryBPMTrainer = "BinaryBPM";
public static PredictorAndArgs BinaryBayesPointMachine = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryBPMTrainer),
MamlArgs = new[] { "norm=no" }
};
public static PredictorAndArgs BinaryBayesPointMachineNoBias = new PredictorAndArgs
{
Trainer = new SubComponent(BinaryBPMTrainer, "bias=-"),
MamlArgs = new[] { "norm=no" },
Tag = "BinaryBPM-NoBias"
};
public static PredictorAndArgs MulticlassNaiveBayesClassifier = new PredictorAndArgs
{
Trainer = new SubComponent("MultiClassNaiveBayes", ""),
BaselineProgress = true,
};
public static PredictorAndArgs FieldAwareFactorizationMachine = new PredictorAndArgs
{
Trainer = new SubComponent("FieldAwareFactorizationMachine", "d=5 shuf- norm-"),
MamlArgs = new[] { "xf=Copy{col=DupFeatures:Features} xf=MinMax{col=Features col=DupFeatures} norm=No", "col[Feature]=DupFeatures" },
BaselineProgress = true
};
}
}
|