private string GetSplitColumn(IChannel ch, IDataView input, ref IDataView output)
{
// The stratification column and/or group column, if they exist at all, must be present at this point.
var schema = input.Schema;
output = input;
// If no stratification column was specified, but we have a group column of type Single, Double or
// Key (contiguous) use it.
string stratificationColumn = null;
if (!string.IsNullOrWhiteSpace(ImplOptions.StratificationColumn))
{
stratificationColumn = ImplOptions.StratificationColumn;
}
else
{
string group = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(ImplOptions.GroupColumn), ImplOptions.GroupColumn, DefaultColumnNames.GroupId);
int index;
if (group != null && schema.TryGetColumnIndex(group, out index))
{
// Check if group column key type with known cardinality.
var type = schema[index].Type;
if (type.GetKeyCount() > 0)
{
stratificationColumn = group;
}
}
}
if (string.IsNullOrEmpty(stratificationColumn))
{
stratificationColumn = "StratificationColumn";
int tmp;
int inc = 0;
while (input.Schema.TryGetColumnIndex(stratificationColumn, out tmp))
{
stratificationColumn = string.Format("StratificationColumn_{0:000}", ++inc);
}
var keyGenArgs = new GenerateNumberTransform.Options();
var col = new GenerateNumberTransform.Column();
col.Name = stratificationColumn;
keyGenArgs.Columns = new[] { col };
output = new GenerateNumberTransform(Host, keyGenArgs, input);
}
else
{
int col;
if (!input.Schema.TryGetColumnIndex(stratificationColumn, out col))
{
throw ch.ExceptUserArg(nameof(Arguments.StratificationColumn), "Column '{0}' does not exist", stratificationColumn);
}
var type = input.Schema[col].Type;
if (!RangeFilter.IsValidRangeFilterColumnType(ch, type))
{
ch.Info("Hashing the stratification column");
var origStratCol = stratificationColumn;
int tmp;
int inc = 0;
while (input.Schema.TryGetColumnIndex(stratificationColumn, out tmp))
{
stratificationColumn = string.Format("{0}_{1:000}", origStratCol, ++inc);
}
output = new HashingEstimator(Host, origStratCol, stratificationColumn, 30).Fit(input).Transform(input);
}
}
return(stratificationColumn);
}
private void RunCore(IChannel ch, string cmd)
{
Host.AssertValue(ch);
IPredictor inputPredictor = null;
if (ImplOptions.ContinueTrain && !TrainUtils.TryLoadPredictor(ch, Host, ImplOptions.InputModelFile, out inputPredictor))
{
ch.Warning("No input model file specified or model file did not contain a predictor. The model state cannot be initialized.");
}
ch.Trace("Constructing data pipeline");
ILegacyDataLoader loader = CreateRawLoader();
// If the per-instance results are requested and there is no name column, add a GenerateNumberTransform.
var preXf = ImplOptions.PreTransforms;
if (!string.IsNullOrEmpty(ImplOptions.OutputDataFile))
{
string name = TrainUtils.MatchNameOrDefaultOrNull(ch, loader.Schema, nameof(ImplOptions.NameColumn), ImplOptions.NameColumn, DefaultColumnNames.Name);
if (name == null)
{
preXf = preXf.Concat(
new[]
{
new KeyValuePair <string, IComponentFactory <IDataView, IDataTransform> >(
"", ComponentFactoryUtils.CreateFromFunction <IDataView, IDataTransform>(
(env, input) =>
{
var args = new GenerateNumberTransform.Options();
args.Columns = new[] { new GenerateNumberTransform.Column()
{
Name = DefaultColumnNames.Name
}, };
args.UseCounter = true;
return(new GenerateNumberTransform(env, args, input));
}))
}).ToArray();
}
}
loader = LegacyCompositeDataLoader.Create(Host, loader, preXf);
ch.Trace("Binding label and features columns");
IDataView pipe = loader;
var stratificationColumn = GetSplitColumn(ch, loader, ref pipe);
var scorer = ImplOptions.Scorer;
var evaluator = ImplOptions.Evaluator;
Func <IDataView> validDataCreator = null;
if (ImplOptions.ValidationFile != null)
{
validDataCreator =
() =>
{
// Fork the command.
var impl = new CrossValidationCommand(this);
return(impl.CreateRawLoader(dataFile: ImplOptions.ValidationFile));
};
}
FoldHelper fold = new FoldHelper(Host, RegistrationName, pipe, stratificationColumn,
ImplOptions, CreateRoleMappedData, ApplyAllTransformsToData, scorer, evaluator,
validDataCreator, ApplyAllTransformsToData, inputPredictor, cmd, loader, !string.IsNullOrEmpty(ImplOptions.OutputDataFile));
var tasks = fold.GetCrossValidationTasks();
var eval = evaluator?.CreateComponent(Host) ??
EvaluateUtils.GetEvaluator(Host, tasks[0].Result.ScoreSchema);
// Print confusion matrix and fold results for each fold.
for (int i = 0; i < tasks.Length; i++)
{
var dict = tasks[i].Result.Metrics;
MetricWriter.PrintWarnings(ch, dict);
eval.PrintFoldResults(ch, dict);
}
// Print the overall results.
if (!TryGetOverallMetrics(tasks.Select(t => t.Result.Metrics).ToArray(), out var overallList))
{
throw ch.Except("No overall metrics found");
}
var overall = eval.GetOverallResults(overallList.ToArray());
MetricWriter.PrintOverallMetrics(Host, ch, ImplOptions.SummaryFilename, overall, ImplOptions.NumFolds);
eval.PrintAdditionalMetrics(ch, tasks.Select(t => t.Result.Metrics).ToArray());
Dictionary <string, IDataView>[] metricValues = tasks.Select(t => t.Result.Metrics).ToArray();
SendTelemetryMetric(metricValues);
// Save the per-instance results.
if (!string.IsNullOrWhiteSpace(ImplOptions.OutputDataFile))
{
var perInstance = EvaluateUtils.ConcatenatePerInstanceDataViews(Host, eval, ImplOptions.CollateMetrics,
ImplOptions.OutputExampleFoldIndex, tasks.Select(t => t.Result.PerInstanceResults).ToArray(), out var variableSizeVectorColumnNames);
if (variableSizeVectorColumnNames.Length > 0)
{
ch.Warning("Detected columns of variable length: {0}. Consider setting collateMetrics- for meaningful per-Folds results.",
string.Join(", ", variableSizeVectorColumnNames));
}
if (ImplOptions.CollateMetrics)
{
ch.Assert(perInstance.Length == 1);
MetricWriter.SavePerInstance(Host, ch, ImplOptions.OutputDataFile, perInstance[0]);
}
else
{
int i = 0;
foreach (var idv in perInstance)
{
MetricWriter.SavePerInstance(Host, ch, ConstructPerFoldName(ImplOptions.OutputDataFile, i), idv);
i++;
}
}
}
}
private void RunCore(IChannel ch, string cmd)
{
Host.AssertValue(ch);
Host.AssertNonEmpty(cmd);
ch.Trace("Constructing trainer");
ITrainer trainer = Args.Trainer.CreateComponent(Host);
IPredictor inputPredictor = null;
if (Args.ContinueTrain && !TrainUtils.TryLoadPredictor(ch, Host, Args.InputModelFile, out inputPredictor))
{
ch.Warning("No input model file specified or model file did not contain a predictor. The model state cannot be initialized.");
}
ch.Trace("Constructing the training pipeline");
IDataView trainPipe = CreateLoader();
var schema = trainPipe.Schema;
string label = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.LabelColumn),
Args.LabelColumn, DefaultColumnNames.Label);
string features = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.FeatureColumn),
Args.FeatureColumn, DefaultColumnNames.Features);
string group = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.GroupColumn),
Args.GroupColumn, DefaultColumnNames.GroupId);
string weight = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.WeightColumn),
Args.WeightColumn, DefaultColumnNames.Weight);
string name = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.NameColumn),
Args.NameColumn, DefaultColumnNames.Name);
TrainUtils.AddNormalizerIfNeeded(Host, ch, trainer, ref trainPipe, features, Args.NormalizeFeatures);
ch.Trace("Binding columns");
var customCols = TrainUtils.CheckAndGenerateCustomColumns(ch, Args.CustomColumn);
var data = new RoleMappedData(trainPipe, label, features, group, weight, name, customCols);
RoleMappedData validData = null;
if (!string.IsNullOrWhiteSpace(Args.ValidationFile))
{
if (!trainer.Info.SupportsValidation)
{
ch.Warning("Ignoring validationFile: Trainer does not accept validation dataset.");
}
else
{
ch.Trace("Constructing the validation pipeline");
IDataView validPipe = CreateRawLoader(dataFile: Args.ValidationFile);
validPipe = ApplyTransformUtils.ApplyAllTransformsToData(Host, trainPipe, validPipe);
validData = new RoleMappedData(validPipe, data.Schema.GetColumnRoleNames());
}
}
// In addition to the training set, some trainers can accept two data sets, validation set and test set,
// in training phase. The major difference between validation set and test set is that training process may
// indirectly use validation set to improve the model but the learned model should totally independent of test set.
// Similar to validation set, the trainer can report the scores computed using test set.
RoleMappedData testDataUsedInTrainer = null;
if (!string.IsNullOrWhiteSpace(Args.TestFile))
{
// In contrast to the if-else block for validation above, we do not throw a warning if test file is provided
// because this is TrainTest command.
if (trainer.Info.SupportsTest)
{
ch.Trace("Constructing the test pipeline");
IDataView testPipeUsedInTrainer = CreateRawLoader(dataFile: Args.TestFile);
testPipeUsedInTrainer = ApplyTransformUtils.ApplyAllTransformsToData(Host, trainPipe, testPipeUsedInTrainer);
testDataUsedInTrainer = new RoleMappedData(testPipeUsedInTrainer, data.Schema.GetColumnRoleNames());
}
}
var predictor = TrainUtils.Train(Host, ch, data, trainer, validData,
Args.Calibrator, Args.MaxCalibrationExamples, Args.CacheData, inputPredictor, testDataUsedInTrainer);
IDataLoader testPipe;
bool hasOutfile = !string.IsNullOrEmpty(Args.OutputModelFile);
var tempFilePath = hasOutfile ? null : Path.GetTempFileName();
using (var file = new SimpleFileHandle(ch, hasOutfile ? Args.OutputModelFile : tempFilePath, true, !hasOutfile))
{
TrainUtils.SaveModel(Host, ch, file, predictor, data, cmd);
ch.Trace("Constructing the testing pipeline");
using (var stream = file.OpenReadStream())
using (var rep = RepositoryReader.Open(stream, ch))
testPipe = LoadLoader(rep, Args.TestFile, true);
}
// Score.
ch.Trace("Scoring and evaluating");
ch.Assert(Args.Scorer == null || Args.Scorer is ICommandLineComponentFactory, "TrainTestCommand should only be used from the command line.");
IDataScorerTransform scorePipe = ScoreUtils.GetScorer(Args.Scorer, predictor, testPipe, features, group, customCols, Host, data.Schema);
// Evaluate.
var evaluator = Args.Evaluator?.CreateComponent(Host) ??
EvaluateUtils.GetEvaluator(Host, scorePipe.Schema);
var dataEval = new RoleMappedData(scorePipe, label, features,
group, weight, name, customCols, opt: true);
var metrics = evaluator.Evaluate(dataEval);
MetricWriter.PrintWarnings(ch, metrics);
evaluator.PrintFoldResults(ch, metrics);
if (!metrics.TryGetValue(MetricKinds.OverallMetrics, out var overall))
{
throw ch.Except("No overall metrics found");
}
overall = evaluator.GetOverallResults(overall);
MetricWriter.PrintOverallMetrics(Host, ch, Args.SummaryFilename, overall, 1);
evaluator.PrintAdditionalMetrics(ch, metrics);
Dictionary <string, IDataView>[] metricValues = { metrics };
SendTelemetryMetric(metricValues);
if (!string.IsNullOrWhiteSpace(Args.OutputDataFile))
{
var perInst = evaluator.GetPerInstanceMetrics(dataEval);
var perInstData = new RoleMappedData(perInst, label, null, group, weight, name, customCols);
var idv = evaluator.GetPerInstanceDataViewToSave(perInstData);
MetricWriter.SavePerInstance(Host, ch, Args.OutputDataFile, idv);
}
}
private void RunCore(IChannel ch, string cmd)
{
Host.AssertValue(ch);
Host.AssertNonEmpty(cmd);
ch.Trace("Constructing trainer");
ITrainer trainer = _trainer.CreateComponent(Host);
IPredictor inputPredictor = null;
if (Args.ContinueTrain && !TrainUtils.TryLoadPredictor(ch, Host, Args.InputModelFile, out inputPredictor))
{
ch.Warning("No input model file specified or model file did not contain a predictor. The model state cannot be initialized.");
}
ch.Trace("Constructing data pipeline");
IDataView view = CreateLoader();
var schema = view.Schema;
var label = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.LabelColumn), _labelColumn, DefaultColumnNames.Label);
var feature = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.FeatureColumn), _featureColumn, DefaultColumnNames.Features);
var group = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.GroupColumn), _groupColumn, DefaultColumnNames.GroupId);
var weight = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.WeightColumn), _weightColumn, DefaultColumnNames.Weight);
var name = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(Arguments.NameColumn), _nameColumn, DefaultColumnNames.Name);
TrainUtils.AddNormalizerIfNeeded(Host, ch, trainer, ref view, feature, Args.NormalizeFeatures);
ch.Trace("Binding columns");
var customCols = TrainUtils.CheckAndGenerateCustomColumns(ch, Args.CustomColumn);
var data = new RoleMappedData(view, label, feature, group, weight, name, customCols);
// REVIEW: Unify the code that creates validation examples in Train, TrainTest and CV commands.
RoleMappedData validData = null;
if (!string.IsNullOrWhiteSpace(Args.ValidationFile))
{
if (!trainer.Info.SupportsValidation)
{
ch.Warning("Ignoring validationFile: Trainer does not accept validation dataset.");
}
else
{
ch.Trace("Constructing the validation pipeline");
IDataView validPipe = CreateRawLoader(dataFile: Args.ValidationFile);
validPipe = ApplyTransformUtils.ApplyAllTransformsToData(Host, view, validPipe);
validData = new RoleMappedData(validPipe, data.Schema.GetColumnRoleNames());
}
}
// In addition to the training set, some trainers can accept two extra data sets, validation set and test set,
// in training phase. The major difference between validation set and test set is that training process may
// indirectly use validation set to improve the model but the learned model should totally independent of test set.
// Similar to validation set, the trainer can report the scores computed using test set.
RoleMappedData testDataUsedInTrainer = null;
if (!string.IsNullOrWhiteSpace(Args.TestFile))
{
// In contrast to the if-else block for validation above, we do not throw a warning if test file is provided
// because this is TrainTest command.
if (trainer.Info.SupportsTest)
{
ch.Trace("Constructing the test pipeline");
IDataView testPipeUsedInTrainer = CreateRawLoader(dataFile: Args.TestFile);
testPipeUsedInTrainer = ApplyTransformUtils.ApplyAllTransformsToData(Host, view, testPipeUsedInTrainer);
testDataUsedInTrainer = new RoleMappedData(testPipeUsedInTrainer, data.Schema.GetColumnRoleNames());
}
}
var predictor = TrainUtils.Train(Host, ch, data, trainer, validData,
Args.Calibrator, Args.MaxCalibrationExamples, Args.CacheData, inputPredictor, testDataUsedInTrainer);
using (var file = Host.CreateOutputFile(Args.OutputModelFile))
TrainUtils.SaveModel(Host, ch, file, predictor, data, cmd);
}
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 = CompositeDataLoader.ApplyTransform(Host, loader, "Scorer", scorer.ToString(),
(env, view) => scorer.CreateComponent(env, view, mapper, trainSchema));
loader = CompositeDataLoader.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.GetMaxMetadataKind(out int colMax, MetadataUtils.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());
}