public void CreateSweepableEstimatorPipelineFromSweepableEstimatorTest()
{
var estimator = SweepableEstimatorFactory.CreateFastForestBinary(new FastForestOption());
var pipeline = estimator.Append(estimator);
pipeline.ToString().Should().Be("FastForestBinary=>FastForestBinary");
}
public void SweepablePipeline_Append_SweepableEstimator_Test()
{
var pipeline = new SweepablePipeline();
var concatOption = new ConcatOption()
{
InputColumnNames = new List <string> {
"a", "b", "c"
}.ToArray(),
OutputColumnName = "a",
};
var lgbmOption = new LgbmOption()
{
FeatureColumnName = "Feature",
LabelColumnName = "Label",
};
// pipeline can append a single sweepable estimator
pipeline = pipeline.Append(SweepableEstimatorFactory.CreateConcatenate(concatOption));
// pipeline can append muliple sweepable estimators.
pipeline = pipeline.Append(SweepableEstimatorFactory.CreateLightGbmBinary(lgbmOption), SweepableEstimatorFactory.CreateConcatenate(concatOption));
// pipeline can append sweepable pipelines mixed with sweepble estimators
pipeline = pipeline.Append(SweepableEstimatorFactory.CreateConcatenate(concatOption), pipeline);
// pipeline can append sweepable pipelines.
pipeline = pipeline.Append(pipeline, pipeline);
Approvals.Verify(JsonSerializer.Serialize(pipeline, _jsonSerializerOptions));
}
public void Execute(GeneratorExecutionContext context)
{
if (context.AdditionalFiles.Where(f => f.Path.Contains("code_gen_flag.json")).First() is AdditionalText text)
{
var json = text.GetText().ToString();
var flags = JsonSerializer.Deserialize <Dictionary <string, bool> >(json);
if (flags.TryGetValue(nameof(SweepableEstimatorFactoryGenerator), out var res) && res == false)
{
return;
}
}
var trainers = context.AdditionalFiles.Where(f => f.Path.Contains("trainer-estimators.json"))
.SelectMany(file => Utils.GetEstimatorsFromJson(file.GetText().ToString()).Estimators, (text, estimator) => (estimator.FunctionName, estimator.EstimatorTypes, estimator.SearchOption))
.SelectMany(union => union.EstimatorTypes.Select(t => (Utils.CreateEstimatorName(union.FunctionName, t), Utils.ToTitleCase(union.SearchOption))))
.ToArray();
var transformers = context.AdditionalFiles.Where(f => f.Path.Contains("transformer-estimators.json"))
.SelectMany(file => Utils.GetEstimatorsFromJson(file.GetText().ToString()).Estimators, (text, estimator) => (estimator.FunctionName, estimator.EstimatorTypes, estimator.SearchOption))
.SelectMany(union => union.EstimatorTypes.Select(t => (Utils.CreateEstimatorName(union.FunctionName, t), Utils.ToTitleCase(union.SearchOption))))
.ToArray();
var code = new SweepableEstimatorFactory()
{
NameSpace = Constant.CodeGeneratorNameSpace,
EstimatorNames = trainers.Concat(transformers),
};
context.AddSource(className + ".cs", SourceText.From(code.TransformText(), Encoding.UTF8));
}
public void CreateSweepableEstimatorPipelineFromSweepableEstimatorAndIEstimatorTest()
{
var context = new MLContext();
var estimator = SweepableEstimatorFactory.CreateFastForestBinary(new FastForestOption());
var pipeline = estimator.Append(context.Transforms.Concatenate("output", "input"));
pipeline.ToString().Should().Be("FastForestBinary=>Unknown");
}
public void AppendIEstimatorToSweepabeEstimatorPipelineTest()
{
var context = new MLContext();
var estimator = context.Transforms.Concatenate("output", "input");
var pipeline = estimator.Append(SweepableEstimatorFactory.CreateFastForestBinary(new FastForestOption()));
pipeline = pipeline.Append(context.Transforms.CopyColumns("output", "input"));
pipeline.ToString().Should().Be("Unknown=>FastForestBinary=>Unknown");
}
public void CreateMultiModelPipelineFromSweepableEstimatorAndMultiClassifiers()
{
var context = new MLContext();
var pipeline = SweepableEstimatorFactory.CreateFastForestBinary(new FastForestOption())
.Append(context.Auto().MultiClassification());
var json = JsonSerializer.Serialize(pipeline, this._jsonSerializerOptions);
Approvals.Verify(json);
}
/// <summary>
/// Create a list of <see cref="SweepableEstimator"/> for featurizing text.
/// </summary>
/// <param name="outputColumnName">output column name.</param>
/// <param name="inputColumnName">input column name.</param>
internal SweepableEstimator[] TextFeaturizer(string outputColumnName, string inputColumnName)
{
var option = new FeaturizeTextOption
{
InputColumnName = inputColumnName,
OutputColumnName = outputColumnName,
};
return(new[] { SweepableEstimatorFactory.CreateFeaturizeText(option) });
}
/// <summary>
/// Create a list of <see cref="SweepableEstimator"/> for multiclass classification.
/// </summary>
/// <param name="labelColumnName">label column name.</param>
/// <param name="featureColumnName">feature column name.</param>
/// <param name="exampleWeightColumnName">example weight column name.</param>
/// <param name="useFastForest">true if use fast forest as available trainer.</param>
/// <param name="useLgbm">true if use lgbm as available trainer.</param>
/// <param name="useFastTree">true if use fast tree as available trainer.</param>
/// <param name="useLbfgs">true if use lbfgs as available trainer.</param>
/// <param name="useSdca">true if use sdca as available trainer.</param>
/// <param name="fastTreeOption">if provided, use it as initial option for fast tree, otherwise the default option will be used.</param>
/// <param name="lgbmOption">if provided, use it as initial option for lgbm, otherwise the default option will be used.</param>
/// <param name="fastForestOption">if provided, use it as initial option for fast forest, otherwise the default option will be used.</param>
/// <param name="lbfgsOption">if provided, use it as initial option for lbfgs, otherwise the default option will be used.</param>
/// <param name="sdcaOption">if provided, use it as initial option for sdca, otherwise the default option will be used.</param>
/// <param name="fastTreeSearchSpace">if provided, use it as search space for fast tree, otherwise the default search space will be used.</param>
/// <param name="lgbmSearchSpace">if provided, use it as search space for lgbm, otherwise the default search space will be used.</param>
/// <param name="fastForestSearchSpace">if provided, use it as search space for fast forest, otherwise the default search space will be used.</param>
/// <param name="lbfgsSearchSpace">if provided, use it as search space for lbfgs, otherwise the default search space will be used.</param>
/// <param name="sdcaSearchSpace">if provided, use it as search space for sdca, otherwise the default search space will be used.</param>
/// <returns></returns>
public SweepableEstimator[] MultiClassification(string labelColumnName = DefaultColumnNames.Label, string featureColumnName = DefaultColumnNames.Features, string exampleWeightColumnName = null, bool useFastForest = true, bool useLgbm = true, bool useFastTree = true, bool useLbfgs = true, bool useSdca = true,
FastTreeOption fastTreeOption = null, LgbmOption lgbmOption = null, FastForestOption fastForestOption = null, LbfgsOption lbfgsOption = null, SdcaOption sdcaOption = null,
SearchSpace <FastTreeOption> fastTreeSearchSpace = null, SearchSpace <LgbmOption> lgbmSearchSpace = null, SearchSpace <FastForestOption> fastForestSearchSpace = null, SearchSpace <LbfgsOption> lbfgsSearchSpace = null, SearchSpace <SdcaOption> sdcaSearchSpace = null)
{
var res = new List <SweepableEstimator>();
if (useFastTree)
{
fastTreeOption = fastTreeOption ?? new FastTreeOption();
fastTreeOption.LabelColumnName = labelColumnName;
fastTreeOption.FeatureColumnName = featureColumnName;
fastTreeOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateFastTreeOva(fastTreeOption, fastTreeSearchSpace ?? new SearchSpace <FastTreeOption>(fastTreeOption)));
}
if (useFastForest)
{
fastForestOption = fastForestOption ?? new FastForestOption();
fastForestOption.LabelColumnName = labelColumnName;
fastForestOption.FeatureColumnName = featureColumnName;
fastForestOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateFastForestOva(fastForestOption, fastForestSearchSpace ?? new SearchSpace <FastForestOption>(fastForestOption)));
}
if (useLgbm)
{
lgbmOption = lgbmOption ?? new LgbmOption();
lgbmOption.LabelColumnName = labelColumnName;
lgbmOption.FeatureColumnName = featureColumnName;
lgbmOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateLightGbmMulti(lgbmOption, lgbmSearchSpace ?? new SearchSpace <LgbmOption>(lgbmOption)));
}
if (useLbfgs)
{
lbfgsOption = lbfgsOption ?? new LbfgsOption();
lbfgsOption.LabelColumnName = labelColumnName;
lbfgsOption.FeatureColumnName = featureColumnName;
lbfgsOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateLbfgsLogisticRegressionOva(lbfgsOption, lbfgsSearchSpace ?? new SearchSpace <LbfgsOption>(lbfgsOption)));
res.Add(SweepableEstimatorFactory.CreateLbfgsMaximumEntropyMulti(lbfgsOption, lbfgsSearchSpace ?? new SearchSpace <LbfgsOption>(lbfgsOption)));
}
if (useSdca)
{
sdcaOption = sdcaOption ?? new SdcaOption();
sdcaOption.LabelColumnName = labelColumnName;
sdcaOption.FeatureColumnName = featureColumnName;
sdcaOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateSdcaMaximumEntropyMulti(sdcaOption, sdcaSearchSpace ?? new SearchSpace <SdcaOption>(sdcaOption)));
res.Add(SweepableEstimatorFactory.CreateSdcaLogisticRegressionOva(sdcaOption, sdcaSearchSpace ?? new SearchSpace <SdcaOption>(sdcaOption)));
}
return(res.ToArray());
}
public void CreateMultiModelPipelineFromSweepableEstimatorPipelineAndMultiClassifiers()
{
var context = new MLContext();
var pipeline = context.Transforms.Concatenate("output", "input")
.Append(SweepableEstimatorFactory.CreateFeaturizeText(new FeaturizeTextOption()))
.Append(context.Auto().MultiClassification());
var json = JsonSerializer.Serialize(pipeline, this._jsonSerializerOptions);
Approvals.Verify(json);
}
internal SweepableEstimator[] Regression(string labelColumnName = DefaultColumnNames.Label, string featureColumnName = DefaultColumnNames.Features, string exampleWeightColumnName = null, bool useFastForest = true, bool useLgbm = true, bool useFastTree = true, bool useLbfgs = true, bool useSdca = true,
FastTreeOption fastTreeOption = null, LgbmOption lgbmOption = null, FastForestOption fastForestOption = null, LbfgsOption lbfgsOption = null, SdcaOption sdcaOption = null,
SearchSpace <FastTreeOption> fastTreeSearchSpace = null, SearchSpace <LgbmOption> lgbmSearchSpace = null, SearchSpace <FastForestOption> fastForestSearchSpace = null, SearchSpace <LbfgsOption> lbfgsSearchSpace = null, SearchSpace <SdcaOption> sdcaSearchSpace = null)
{
var res = new List <SweepableEstimator>();
if (useFastTree)
{
fastTreeOption = fastTreeOption ?? new FastTreeOption();
fastTreeOption.LabelColumnName = labelColumnName;
fastTreeOption.FeatureColumnName = featureColumnName;
fastTreeOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateFastTreeRegression(fastTreeOption, fastTreeSearchSpace ?? new SearchSpace <FastTreeOption>()));
res.Add(SweepableEstimatorFactory.CreateFastTreeTweedieRegression(fastTreeOption, fastTreeSearchSpace ?? new SearchSpace <FastTreeOption>()));
}
if (useFastForest)
{
fastForestOption = fastForestOption ?? new FastForestOption();
fastForestOption.LabelColumnName = labelColumnName;
fastForestOption.FeatureColumnName = featureColumnName;
fastForestOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateFastForestRegression(fastForestOption, fastForestSearchSpace ?? new SearchSpace <FastForestOption>()));
}
if (useLgbm)
{
lgbmOption = lgbmOption ?? new LgbmOption();
lgbmOption.LabelColumnName = labelColumnName;
lgbmOption.FeatureColumnName = featureColumnName;
lgbmOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateLightGbmRegression(lgbmOption, lgbmSearchSpace ?? new SearchSpace <LgbmOption>()));
}
if (useLbfgs)
{
lbfgsOption = lbfgsOption ?? new LbfgsOption();
lbfgsOption.LabelColumnName = labelColumnName;
lbfgsOption.FeatureColumnName = featureColumnName;
lbfgsOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateLbfgsPoissonRegressionRegression(lbfgsOption, lbfgsSearchSpace ?? new SearchSpace <LbfgsOption>()));
}
if (useSdca)
{
sdcaOption = sdcaOption ?? new SdcaOption();
sdcaOption.LabelColumnName = labelColumnName;
sdcaOption.FeatureColumnName = featureColumnName;
sdcaOption.ExampleWeightColumnName = exampleWeightColumnName;
res.Add(SweepableEstimatorFactory.CreateSdcaRegression(sdcaOption, sdcaSearchSpace ?? new SearchSpace <SdcaOption>()));
}
return(res.ToArray());
}
/// <summary>
/// Create a list of <see cref="SweepableEstimator"/> for featurizing catalog columns.
/// </summary>
/// <param name="outputColumnNames">output column names.</param>
/// <param name="inputColumnNames">input column names.</param>
internal SweepableEstimator[] CatalogFeaturizer(string[] outputColumnNames, string[] inputColumnNames)
{
Contracts.Check(outputColumnNames.Count() == inputColumnNames.Count() && outputColumnNames.Count() > 0, "outputColumnNames and inputColumnNames must have the same length and greater than 0");
var option = new OneHotOption
{
InputColumnNames = inputColumnNames,
OutputColumnNames = outputColumnNames,
};
return(new SweepableEstimator[] { SweepableEstimatorFactory.CreateOneHotEncoding(option), SweepableEstimatorFactory.CreateOneHotHashEncoding(option) });
}
private SweepableEstimatorPipeline CreateSweepbaleEstimatorPipeline()
{
var concat = SweepableEstimatorFactory.CreateConcatenate(new ConcatOption());
var replaceMissingValue = SweepableEstimatorFactory.CreateReplaceMissingValues(new ReplaceMissingValueOption());
var oneHot = SweepableEstimatorFactory.CreateOneHotEncoding(new OneHotOption());
var lightGbm = SweepableEstimatorFactory.CreateLightGbmBinary(new LgbmOption());
var fastTree = SweepableEstimatorFactory.CreateFastTreeBinary(new FastTreeOption());
var pipeline = new SweepableEstimatorPipeline(new SweepableEstimator[] { concat, replaceMissingValue, oneHot, lightGbm, fastTree });
return(pipeline);
}
/// <summary>
/// Create a list of <see cref="SweepableEstimator"/> for featurizing numeric columns.
/// </summary>
/// <param name="outputColumnNames">output column names.</param>
/// <param name="inputColumnNames">input column names.</param>
internal SweepableEstimator[] NumericFeaturizer(string[] outputColumnNames, string[] inputColumnNames)
{
Contracts.CheckValue(inputColumnNames, nameof(inputColumnNames));
Contracts.CheckValue(outputColumnNames, nameof(outputColumnNames));
Contracts.Check(outputColumnNames.Count() == inputColumnNames.Count() && outputColumnNames.Count() > 0, "outputColumnNames and inputColumnNames must have the same length and greater than 0");
var replaceMissingValueOption = new ReplaceMissingValueOption
{
InputColumnNames = inputColumnNames,
OutputColumnNames = outputColumnNames,
};
return(new[] { SweepableEstimatorFactory.CreateReplaceMissingValues(replaceMissingValueOption) });
}
private MultiModelPipeline CreateMultiModelPipeline()
{
var concat = SweepableEstimatorFactory.CreateConcatenate(new ConcatOption());
var replaceMissingValue = SweepableEstimatorFactory.CreateReplaceMissingValues(new ReplaceMissingValueOption());
var oneHot = SweepableEstimatorFactory.CreateOneHotEncoding(new OneHotOption());
var lightGbm = SweepableEstimatorFactory.CreateLightGbmBinary(new LgbmOption());
var fastTree = SweepableEstimatorFactory.CreateFastTreeBinary(new FastTreeOption());
var pipeline = new MultiModelPipeline();
pipeline = pipeline.AppendOrSkip(replaceMissingValue + replaceMissingValue * oneHot);
pipeline = pipeline.AppendOrSkip(concat);
pipeline = pipeline.Append(lightGbm + fastTree);
return(pipeline);
}
/// <summary>
/// Create a single featurize pipeline according to <paramref name="columnInformation"/>. This function will collect all columns in <paramref name="columnInformation"/>,
/// featurizing them using <see cref="CatalogFeaturizer(string[], string[])"/>, <see cref="NumericFeaturizer(string[], string[])"/> or <see cref="TextFeaturizer(string, string)"/>. And combine
/// them into a single feature column as output.
/// </summary>
/// <param name="data">input data.</param>
/// <param name="columnInformation">column information.</param>
/// <param name="outputColumnName">output feature column.</param>
/// <returns>A <see cref="MultiModelPipeline"/> for featurization.</returns>
public MultiModelPipeline Featurizer(IDataView data, ColumnInformation columnInformation, string outputColumnName = "Features")
{
Contracts.CheckValue(data, nameof(data));
Contracts.CheckValue(columnInformation, nameof(columnInformation));
var columnPurposes = PurposeInference.InferPurposes(this._context, data, columnInformation);
var textFeatures = columnPurposes.Where(c => c.Purpose == ColumnPurpose.TextFeature);
var numericFeatures = columnPurposes.Where(c => c.Purpose == ColumnPurpose.NumericFeature);
var catalogFeatures = columnPurposes.Where(c => c.Purpose == ColumnPurpose.CategoricalFeature);
var textFeatureColumnNames = textFeatures.Select(c => data.Schema[c.ColumnIndex].Name).ToArray();
var numericFeatureColumnNames = numericFeatures.Select(c => data.Schema[c.ColumnIndex].Name).ToArray();
var catalogFeatureColumnNames = catalogFeatures.Select(c => data.Schema[c.ColumnIndex].Name).ToArray();
var pipeline = new MultiModelPipeline();
if (numericFeatureColumnNames.Length > 0)
{
pipeline = pipeline.Append(this.NumericFeaturizer(numericFeatureColumnNames, numericFeatureColumnNames));
}
if (catalogFeatureColumnNames.Length > 0)
{
pipeline = pipeline.Append(this.CatalogFeaturizer(catalogFeatureColumnNames, catalogFeatureColumnNames));
}
foreach (var textColumn in textFeatureColumnNames)
{
pipeline = pipeline.Append(this.TextFeaturizer(textColumn, textColumn));
}
var option = new ConcatOption
{
InputColumnNames = textFeatureColumnNames.Concat(numericFeatureColumnNames).Concat(catalogFeatureColumnNames).ToArray(),
OutputColumnName = outputColumnName,
};
if (option.InputColumnNames.Length > 0)
{
pipeline = pipeline.Append(SweepableEstimatorFactory.CreateConcatenate(option));
}
return(pipeline);
}
