public void TestCrossValidationMacroWithStratification()
{
var dataPath = GetDataPath(@"breast-cancer.txt");
using (var env = new TlcEnvironment(42))
{
var subGraph = env.CreateExperiment();
var nop = new ML.Transforms.NoOperation();
var nopOutput = subGraph.Add(nop);
var learnerInput = new ML.Trainers.StochasticDualCoordinateAscentBinaryClassifier
{
TrainingData = nopOutput.OutputData,
NumThreads = 1
};
var learnerOutput = subGraph.Add(learnerInput);
var modelCombine = new ML.Transforms.ManyHeterogeneousModelCombiner
{
TransformModels = new ArrayVar <ITransformModel>(nopOutput.Model),
PredictorModel = learnerOutput.PredictorModel
};
var modelCombineOutput = subGraph.Add(modelCombine);
var experiment = env.CreateExperiment();
var importInput = new ML.Data.TextLoader(dataPath);
importInput.Arguments.Column = new ML.Data.TextLoaderColumn[]
{
new ML.Data.TextLoaderColumn {
Name = "Label", Source = new[] { new ML.Data.TextLoaderRange(0) }
},
new ML.Data.TextLoaderColumn {
Name = "Strat", Source = new[] { new ML.Data.TextLoaderRange(1) }
},
new ML.Data.TextLoaderColumn {
Name = "Features", Source = new[] { new ML.Data.TextLoaderRange(2, 9) }
}
};
var importOutput = experiment.Add(importInput);
var crossValidate = new ML.Models.CrossValidator
{
Data = importOutput.Data,
Nodes = subGraph,
TransformModel = null,
StratificationColumn = "Strat"
};
crossValidate.Inputs.Data = nop.Data;
crossValidate.Outputs.PredictorModel = modelCombineOutput.PredictorModel;
var crossValidateOutput = experiment.Add(crossValidate);
experiment.Compile();
experiment.SetInput(importInput.InputFile, new SimpleFileHandle(env, dataPath, false, false));
experiment.Run();
var data = experiment.GetOutput(crossValidateOutput.OverallMetrics);
var schema = data.Schema;
var b = schema.TryGetColumnIndex("AUC", out int metricCol);
Assert.True(b);
b = schema.TryGetColumnIndex("Fold Index", out int foldCol);
Assert.True(b);
using (var cursor = data.GetRowCursor(col => col == metricCol || col == foldCol))
{
var getter = cursor.GetGetter <double>(metricCol);
var foldGetter = cursor.GetGetter <DvText>(foldCol);
DvText fold = default;
// Get the verage.
b = cursor.MoveNext();
Assert.True(b);
double avg = 0;
getter(ref avg);
foldGetter(ref fold);
Assert.True(fold.EqualsStr("Average"));
// Get the standard deviation.
b = cursor.MoveNext();
Assert.True(b);
double stdev = 0;
getter(ref stdev);
foldGetter(ref fold);
Assert.True(fold.EqualsStr("Standard Deviation"));
Assert.Equal(0.00485, stdev, 5);
double sum = 0;
double val = 0;
for (int f = 0; f < 2; f++)
{
b = cursor.MoveNext();
Assert.True(b);
getter(ref val);
foldGetter(ref fold);
sum += val;
Assert.True(fold.EqualsStr("Fold " + f));
}
Assert.Equal(avg, sum / 2);
b = cursor.MoveNext();
Assert.False(b);
}
}
}
public void TestCrossValidationMacroWithMultiClass()
{
var dataPath = GetDataPath(@"Train-Tiny-28x28.txt");
using (var env = new TlcEnvironment(42))
{
var subGraph = env.CreateExperiment();
var nop = new ML.Transforms.NoOperation();
var nopOutput = subGraph.Add(nop);
var learnerInput = new ML.Trainers.StochasticDualCoordinateAscentClassifier
{
TrainingData = nopOutput.OutputData,
NumThreads = 1
};
var learnerOutput = subGraph.Add(learnerInput);
var modelCombine = new ML.Transforms.ManyHeterogeneousModelCombiner
{
TransformModels = new ArrayVar <ITransformModel>(nopOutput.Model),
PredictorModel = learnerOutput.PredictorModel
};
var modelCombineOutput = subGraph.Add(modelCombine);
var experiment = env.CreateExperiment();
var importInput = new ML.Data.TextLoader(dataPath);
var importOutput = experiment.Add(importInput);
var crossValidate = new ML.Models.CrossValidator
{
Data = importOutput.Data,
Nodes = subGraph,
Kind = ML.Models.MacroUtilsTrainerKinds.SignatureMultiClassClassifierTrainer,
TransformModel = null
};
crossValidate.Inputs.Data = nop.Data;
crossValidate.Outputs.PredictorModel = modelCombineOutput.PredictorModel;
var crossValidateOutput = experiment.Add(crossValidate);
experiment.Compile();
importInput.SetInput(env, experiment);
experiment.Run();
var data = experiment.GetOutput(crossValidateOutput.OverallMetrics);
var schema = data.Schema;
var b = schema.TryGetColumnIndex("Accuracy(micro-avg)", out int metricCol);
Assert.True(b);
b = schema.TryGetColumnIndex("Fold Index", out int foldCol);
Assert.True(b);
using (var cursor = data.GetRowCursor(col => col == metricCol || col == foldCol))
{
var getter = cursor.GetGetter <double>(metricCol);
var foldGetter = cursor.GetGetter <DvText>(foldCol);
DvText fold = default;
// Get the verage.
b = cursor.MoveNext();
Assert.True(b);
double avg = 0;
getter(ref avg);
foldGetter(ref fold);
Assert.True(fold.EqualsStr("Average"));
// Get the standard deviation.
b = cursor.MoveNext();
Assert.True(b);
double stdev = 0;
getter(ref stdev);
foldGetter(ref fold);
Assert.True(fold.EqualsStr("Standard Deviation"));
Assert.Equal(0.025, stdev, 3);
double sum = 0;
double val = 0;
for (int f = 0; f < 2; f++)
{
b = cursor.MoveNext();
Assert.True(b);
getter(ref val);
foldGetter(ref fold);
sum += val;
Assert.True(fold.EqualsStr("Fold " + f));
}
Assert.Equal(avg, sum / 2);
b = cursor.MoveNext();
Assert.False(b);
}
var confusion = experiment.GetOutput(crossValidateOutput.ConfusionMatrix);
schema = confusion.Schema;
b = schema.TryGetColumnIndex("Count", out int countCol);
Assert.True(b);
b = schema.TryGetColumnIndex("Fold Index", out foldCol);
Assert.True(b);
var type = schema.GetMetadataTypeOrNull(MetadataUtils.Kinds.SlotNames, countCol);
Assert.True(type != null && type.ItemType.IsText && type.VectorSize == 10);
var slotNames = default(VBuffer <DvText>);
schema.GetMetadata(MetadataUtils.Kinds.SlotNames, countCol, ref slotNames);
Assert.True(slotNames.Values.Select((s, i) => s.EqualsStr(i.ToString())).All(x => x));
using (var curs = confusion.GetRowCursor(col => true))
{
var countGetter = curs.GetGetter <VBuffer <double> >(countCol);
var foldGetter = curs.GetGetter <DvText>(foldCol);
var confCount = default(VBuffer <double>);
var foldIndex = default(DvText);
int rowCount = 0;
var foldCur = "Fold 0";
while (curs.MoveNext())
{
countGetter(ref confCount);
foldGetter(ref foldIndex);
rowCount++;
Assert.True(foldIndex.EqualsStr(foldCur));
if (rowCount == 10)
{
rowCount = 0;
foldCur = "Fold 1";
}
}
Assert.Equal(0, rowCount);
}
var warnings = experiment.GetOutput(crossValidateOutput.Warnings);
using (var cursor = warnings.GetRowCursor(col => true))
Assert.False(cursor.MoveNext());
}
}
public void TestCrossValidationMacroMultiClassWithWarnings()
{
var dataPath = GetDataPath(@"Train-Tiny-28x28.txt");
using (var env = new TlcEnvironment(42))
{
var subGraph = env.CreateExperiment();
var nop = new ML.Transforms.NoOperation();
var nopOutput = subGraph.Add(nop);
var learnerInput = new ML.Trainers.LogisticRegressionClassifier
{
TrainingData = nopOutput.OutputData,
NumThreads = 1
};
var learnerOutput = subGraph.Add(learnerInput);
var experiment = env.CreateExperiment();
var importInput = new ML.Data.TextLoader(dataPath);
var importOutput = experiment.Add(importInput);
var filter = new ML.Transforms.RowRangeFilter();
filter.Data = importOutput.Data;
filter.Column = "Label";
filter.Min = 0;
filter.Max = 5;
var filterOutput = experiment.Add(filter);
var term = new ML.Transforms.TextToKeyConverter();
term.Column = new[]
{
new ML.Transforms.TermTransformColumn()
{
Source = "Label", Name = "Strat", Sort = ML.Transforms.TermTransformSortOrder.Value
}
};
term.Data = filterOutput.OutputData;
var termOutput = experiment.Add(term);
var crossValidate = new ML.Models.CrossValidator
{
Data = termOutput.OutputData,
Nodes = subGraph,
Kind = ML.Models.MacroUtilsTrainerKinds.SignatureMultiClassClassifierTrainer,
TransformModel = null,
StratificationColumn = "Strat"
};
crossValidate.Inputs.Data = nop.Data;
crossValidate.Outputs.PredictorModel = learnerOutput.PredictorModel;
var crossValidateOutput = experiment.Add(crossValidate);
experiment.Compile();
importInput.SetInput(env, experiment);
experiment.Run();
var warnings = experiment.GetOutput(crossValidateOutput.Warnings);
var schema = warnings.Schema;
var b = schema.TryGetColumnIndex("WarningText", out int warningCol);
Assert.True(b);
using (var cursor = warnings.GetRowCursor(col => col == warningCol))
{
var getter = cursor.GetGetter <DvText>(warningCol);
b = cursor.MoveNext();
Assert.True(b);
var warning = default(DvText);
getter(ref warning);
Assert.Contains("test instances with class values not seen in the training set.", warning.ToString());
b = cursor.MoveNext();
Assert.True(b);
getter(ref warning);
Assert.Contains("Detected columns of variable length: SortedScores, SortedClasses", warning.ToString());
b = cursor.MoveNext();
Assert.False(b);
}
}
}
public void TestCrossValidationMacro()
{
var dataPath = GetDataPath(TestDatasets.winequality.trainFilename);
using (var env = new TlcEnvironment(42))
{
var subGraph = env.CreateExperiment();
var nop = new ML.Transforms.NoOperation();
var nopOutput = subGraph.Add(nop);
var generate = new ML.Transforms.RandomNumberGenerator();
generate.Column = new[] { new ML.Transforms.GenerateNumberTransformColumn()
{
Name = "Weight1"
} };
generate.Data = nopOutput.OutputData;
var generateOutput = subGraph.Add(generate);
var learnerInput = new ML.Trainers.PoissonRegressor
{
TrainingData = generateOutput.OutputData,
NumThreads = 1,
WeightColumn = "Weight1"
};
var learnerOutput = subGraph.Add(learnerInput);
var modelCombine = new ML.Transforms.ManyHeterogeneousModelCombiner
{
TransformModels = new ArrayVar <ITransformModel>(nopOutput.Model, generateOutput.Model),
PredictorModel = learnerOutput.PredictorModel
};
var modelCombineOutput = subGraph.Add(modelCombine);
var experiment = env.CreateExperiment();
var importInput = new ML.Data.TextLoader(dataPath)
{
Arguments = new TextLoaderArguments
{
Separator = new[] { ';' },
HasHeader = true,
Column = new[]
{
new TextLoaderColumn()
{
Name = "Label",
Source = new [] { new TextLoaderRange(11) },
Type = ML.Data.DataKind.Num
},
new TextLoaderColumn()
{
Name = "Features",
Source = new [] { new TextLoaderRange(0, 10) },
Type = ML.Data.DataKind.Num
}
}
}
};
var importOutput = experiment.Add(importInput);
var crossValidate = new ML.Models.CrossValidator
{
Data = importOutput.Data,
Nodes = subGraph,
Kind = ML.Models.MacroUtilsTrainerKinds.SignatureRegressorTrainer,
TransformModel = null,
WeightColumn = "Weight1"
};
crossValidate.Inputs.Data = nop.Data;
crossValidate.Outputs.PredictorModel = modelCombineOutput.PredictorModel;
var crossValidateOutput = experiment.Add(crossValidate);
experiment.Compile();
importInput.SetInput(env, experiment);
experiment.Run();
var data = experiment.GetOutput(crossValidateOutput.OverallMetrics);
var schema = data.Schema;
var b = schema.TryGetColumnIndex("L1(avg)", out int metricCol);
Assert.True(b);
b = schema.TryGetColumnIndex("Fold Index", out int foldCol);
Assert.True(b);
b = schema.TryGetColumnIndex("IsWeighted", out int isWeightedCol);
using (var cursor = data.GetRowCursor(col => col == metricCol || col == foldCol || col == isWeightedCol))
{
var getter = cursor.GetGetter <double>(metricCol);
var foldGetter = cursor.GetGetter <DvText>(foldCol);
var isWeightedGetter = cursor.GetGetter <DvBool>(isWeightedCol);
DvText fold = default;
DvBool isWeighted = default;
double avg = 0;
double weightedAvg = 0;
for (int w = 0; w < 2; w++)
{
// Get the average.
b = cursor.MoveNext();
Assert.True(b);
if (w == 1)
{
getter(ref weightedAvg);
}
else
{
getter(ref avg);
}
foldGetter(ref fold);
Assert.True(fold.EqualsStr("Average"));
isWeightedGetter(ref isWeighted);
Assert.True(isWeighted.IsTrue == (w == 1));
// Get the standard deviation.
b = cursor.MoveNext();
Assert.True(b);
double stdev = 0;
getter(ref stdev);
foldGetter(ref fold);
Assert.True(fold.EqualsStr("Standard Deviation"));
if (w == 1)
{
Assert.Equal(0.002827, stdev, 6);
}
else
{
Assert.Equal(0.002376, stdev, 6);
}
isWeightedGetter(ref isWeighted);
Assert.True(isWeighted.IsTrue == (w == 1));
}
double sum = 0;
double weightedSum = 0;
for (int f = 0; f < 2; f++)
{
for (int w = 0; w < 2; w++)
{
b = cursor.MoveNext();
Assert.True(b);
double val = 0;
getter(ref val);
foldGetter(ref fold);
if (w == 1)
{
weightedSum += val;
}
else
{
sum += val;
}
Assert.True(fold.EqualsStr("Fold " + f));
isWeightedGetter(ref isWeighted);
Assert.True(isWeighted.IsTrue == (w == 1));
}
}
Assert.Equal(weightedAvg, weightedSum / 2);
Assert.Equal(avg, sum / 2);
b = cursor.MoveNext();
Assert.False(b);
}
}
}
