// Returns true if a normalizer was added.
public static bool AddNormalizerIfNeeded(IHostEnvironment env, IChannel ch, ITrainer trainer, ref IDataView view, string featureColumn, NormalizeOption autoNorm)
{
Contracts.CheckValue(env, nameof(env));
env.CheckValue(ch, nameof(ch));
ch.CheckValue(trainer, nameof(trainer));
ch.CheckValue(view, nameof(view));
ch.CheckValueOrNull(featureColumn);
ch.CheckUserArg(Enum.IsDefined(typeof(NormalizeOption), autoNorm), nameof(TrainCommand.Arguments.NormalizeFeatures),
"Normalize option is invalid. Specify one of 'norm=No', 'norm=Warn', 'norm=Auto', or 'norm=Yes'.");
if (autoNorm == NormalizeOption.No)
{
ch.Info("Not adding a normalizer.");
return(false);
}
if (string.IsNullOrEmpty(featureColumn))
{
return(false);
}
int featCol;
var schema = view.Schema;
if (schema.TryGetColumnIndex(featureColumn, out featCol))
{
if (autoNorm != NormalizeOption.Yes)
{
if (!trainer.Info.NeedNormalization || schema[featCol].IsNormalized())
{
ch.Info("Not adding a normalizer.");
return(false);
}
if (autoNorm == NormalizeOption.Warn)
{
ch.Warning("A normalizer is needed for this trainer. Either add a normalizing transform or use the 'norm=Auto', 'norm=Yes' or 'norm=No' options.");
return(false);
}
}
ch.Info("Automatically adding a MinMax normalization transform, use 'norm=Warn' or 'norm=No' to turn this behavior off.");
IDataView ApplyNormalizer(IHostEnvironment innerEnv, IDataView input)
=> NormalizeTransform.CreateMinMaxNormalizer(innerEnv, input, featureColumn);
if (view is ILegacyDataLoader loader)
{
view = LegacyCompositeDataLoader.ApplyTransform(env, loader, tag: null, creationArgs: null, ApplyNormalizer);
}
else
{
view = ApplyNormalizer(env, view);
}
return(true);
}
return(false);
}
private void RunCore(IChannel ch)
{
Host.AssertValue(ch);
ch.Trace("Creating loader");
LoadModelObjects(ch, true, out var predictor, true, out var trainSchema, out var loader);
ch.AssertValue(predictor);
ch.AssertValueOrNull(trainSchema);
ch.AssertValue(loader);
ch.Trace("Creating pipeline");
var scorer = ImplOptions.Scorer;
ch.Assert(scorer == null || scorer is ICommandLineComponentFactory, "ScoreCommand should only be used from the command line.");
var bindable = ScoreUtils.GetSchemaBindableMapper(Host, predictor, scorerFactorySettings: scorer as ICommandLineComponentFactory);
ch.AssertValue(bindable);
// REVIEW: We probably ought to prefer role mappings from the training schema.
string feat = TrainUtils.MatchNameOrDefaultOrNull(ch, loader.Schema,
nameof(ImplOptions.FeatureColumn), ImplOptions.FeatureColumn, DefaultColumnNames.Features);
string group = TrainUtils.MatchNameOrDefaultOrNull(ch, loader.Schema,
nameof(ImplOptions.GroupColumn), ImplOptions.GroupColumn, DefaultColumnNames.GroupId);
var customCols = TrainUtils.CheckAndGenerateCustomColumns(ch, ImplOptions.CustomColumns);
var schema = new RoleMappedSchema(loader.Schema, label: null, feature: feat, group: group, custom: customCols, opt: true);
var mapper = bindable.Bind(Host, schema);
if (scorer == null)
{
scorer = ScoreUtils.GetScorerComponent(Host, mapper);
}
loader = LegacyCompositeDataLoader.ApplyTransform(Host, loader, "Scorer", scorer.ToString(),
(env, view) => scorer.CreateComponent(env, view, mapper, trainSchema));
loader = LegacyCompositeDataLoader.Create(Host, loader, ImplOptions.PostTransform);
if (!string.IsNullOrWhiteSpace(ImplOptions.OutputModelFile))
{
ch.Trace("Saving the data pipe");
SaveLoader(loader, ImplOptions.OutputModelFile);
}
ch.Trace("Creating saver");
IDataSaver writer;
if (ImplOptions.Saver == null)
{
var ext = Path.GetExtension(ImplOptions.OutputDataFile);
var isText = ext == ".txt" || ext == ".tlc";
if (isText)
{
writer = new TextSaver(Host, new TextSaver.Arguments());
}
else
{
writer = new BinarySaver(Host, new BinarySaver.Arguments());
}
}
else
{
writer = ImplOptions.Saver.CreateComponent(Host);
}
ch.Assert(writer != null);
var outputIsBinary = writer is BinaryWriter;
bool outputAllColumns =
ImplOptions.OutputAllColumns == true ||
(ImplOptions.OutputAllColumns == null && Utils.Size(ImplOptions.OutputColumns) == 0 && outputIsBinary);
bool outputNamesAndLabels =
ImplOptions.OutputAllColumns == true || Utils.Size(ImplOptions.OutputColumns) == 0;
if (ImplOptions.OutputAllColumns == true && Utils.Size(ImplOptions.OutputColumns) != 0)
{
ch.Warning(nameof(ImplOptions.OutputAllColumns) + "=+ always writes all columns irrespective of " + nameof(ImplOptions.OutputColumns) + " specified.");
}
if (!outputAllColumns && Utils.Size(ImplOptions.OutputColumns) != 0)
{
foreach (var outCol in ImplOptions.OutputColumns)
{
if (!loader.Schema.TryGetColumnIndex(outCol, out int dummyColIndex))
{
throw ch.ExceptUserArg(nameof(Arguments.OutputColumns), "Column '{0}' not found.", outCol);
}
}
}
uint maxScoreId = 0;
if (!outputAllColumns)
{
maxScoreId = loader.Schema.GetMaxAnnotationKind(out int colMax, AnnotationUtils.Kinds.ScoreColumnSetId);
}
ch.Assert(outputAllColumns || maxScoreId > 0); // score set IDs are one-based
var cols = new List <int>();
for (int i = 0; i < loader.Schema.Count; i++)
{
if (!ImplOptions.KeepHidden && loader.Schema[i].IsHidden)
{
continue;
}
if (!(outputAllColumns || ShouldAddColumn(loader.Schema, i, maxScoreId, outputNamesAndLabels)))
{
continue;
}
var type = loader.Schema[i].Type;
if (writer.IsColumnSavable(type))
{
cols.Add(i);
}
else
{
ch.Warning("The column '{0}' will not be written as it has unsavable column type.",
loader.Schema[i].Name);
}
}
ch.Check(cols.Count > 0, "No valid columns to save");
ch.Trace("Scoring and saving data");
using (var file = Host.CreateOutputFile(ImplOptions.OutputDataFile))
using (var stream = file.CreateWriteStream())
writer.SaveData(stream, loader, cols.ToArray());
}
private FoldResult RunFold(int fold)
{
var host = GetHost();
host.Assert(0 <= fold && fold <= _numFolds);
// REVIEW: Make channels buffered in multi-threaded environments.
using (var ch = host.Start($"Fold {fold}"))
{
ch.Trace("Constructing trainer");
ITrainer trainer = _trainer.CreateComponent(host);
// Train pipe.
var trainFilter = new RangeFilter.Options();
trainFilter.Column = _splitColumn;
trainFilter.Min = (Double)fold / _numFolds;
trainFilter.Max = (Double)(fold + 1) / _numFolds;
trainFilter.Complement = true;
IDataView trainPipe = new RangeFilter(host, trainFilter, _inputDataView);
trainPipe = new OpaqueDataView(trainPipe);
var trainData = _createExamples(host, ch, trainPipe, trainer);
// Test pipe.
var testFilter = new RangeFilter.Options();
testFilter.Column = trainFilter.Column;
testFilter.Min = trainFilter.Min;
testFilter.Max = trainFilter.Max;
ch.Assert(!testFilter.Complement);
IDataView testPipe = new RangeFilter(host, testFilter, _inputDataView);
testPipe = new OpaqueDataView(testPipe);
var testData = _applyTransformsToTestData(host, ch, testPipe, trainData, trainPipe);
// Validation pipe and examples.
RoleMappedData validData = null;
if (_getValidationDataView != null)
{
ch.Assert(_applyTransformsToValidationData != null);
if (!trainer.Info.SupportsValidation)
{
ch.Warning("Trainer does not accept validation dataset.");
}
else
{
ch.Trace("Constructing the validation pipeline");
IDataView validLoader = _getValidationDataView();
var validPipe = ApplyTransformUtils.ApplyAllTransformsToData(host, _inputDataView, validLoader);
validPipe = new OpaqueDataView(validPipe);
validData = _applyTransformsToValidationData(host, ch, validPipe, trainData, trainPipe);
}
}
// Train.
var predictor = TrainUtils.Train(host, ch, trainData, trainer, validData,
_calibrator, _maxCalibrationExamples, _cacheData, _inputPredictor);
// Score.
ch.Trace("Scoring and evaluating");
ch.Assert(_scorer == null || _scorer is ICommandLineComponentFactory, "CrossValidationCommand should only be used from the command line.");
var bindable = ScoreUtils.GetSchemaBindableMapper(host, predictor, scorerFactorySettings: _scorer as ICommandLineComponentFactory);
ch.AssertValue(bindable);
var mapper = bindable.Bind(host, testData.Schema);
var scorerComp = _scorer ?? ScoreUtils.GetScorerComponent(host, mapper);
IDataScorerTransform scorePipe = scorerComp.CreateComponent(host, testData.Data, mapper, trainData.Schema);
// Save per-fold model.
string modelFileName = ConstructPerFoldName(_outputModelFile, fold);
if (modelFileName != null && _loader != null)
{
using (var file = host.CreateOutputFile(modelFileName))
{
var rmd = new RoleMappedData(
LegacyCompositeDataLoader.ApplyTransform(host, _loader, null, null,
(e, newSource) => ApplyTransformUtils.ApplyAllTransformsToData(e, trainData.Data, newSource)),
trainData.Schema.GetColumnRoleNames());
TrainUtils.SaveModel(host, ch, file, predictor, rmd, _cmd);
}
}
// Evaluate.
var eval = _evaluator?.CreateComponent(host) ??
EvaluateUtils.GetEvaluator(host, scorePipe.Schema);
// Note that this doesn't require the provided columns to exist (because of the "opt" parameter).
// We don't normally expect the scorer to drop columns, but if it does, we should not require
// all the columns in the test pipeline to still be present.
var dataEval = new RoleMappedData(scorePipe, testData.Schema.GetColumnRoleNames(), opt: true);
var dict = eval.Evaluate(dataEval);
RoleMappedData perInstance = null;
if (_savePerInstance)
{
var perInst = eval.GetPerInstanceMetrics(dataEval);
perInstance = new RoleMappedData(perInst, dataEval.Schema.GetColumnRoleNames(), opt: true);
}
return(new FoldResult(dict, dataEval.Schema.Schema, perInstance, trainData.Schema));
}
}
