protected override INearestNeighborsPredictor Train(RoleMappedData data)
{
int count;
data.CheckMultiClassLabel(out count);
return(base.Train(data));
}
public override void Train(RoleMappedData data)
{
Host.CheckValue(data, nameof(data));
data.CheckFeatureFloatVector();
int count;
data.CheckMultiClassLabel(out count);
Host.Assert(count > 0);
using (var ch = Host.Start("Training"))
{
_pred = TrainCore(ch, data, count);
ch.Check(_pred != null, "Training did not result in a predictor");
ch.Done();
}
}
TVectorPredictor Train(RoleMappedData data)
{
Contracts.CheckValue(data, "data");
data.CheckFeatureFloatVector();
int count;
data.CheckMultiClassLabel(out count);
using (var ch = Host.Start("Training"))
{
// Train one-vs-all models.
_predictors = new TVectorPredictor[1];
for (int i = 0; i < _predictors.Length; i++)
{
ch.Info("Training learner {0}", i);
Contracts.CheckValue(_args.predictorType, "predictorType", "Must specify a base learner type");
TScalarTrainer trainer;
if (_trainer != null)
{
trainer = _trainer;
}
else
{
var temp = ScikitSubComponent <ITrainer, SignatureBinaryClassifierTrainer> .AsSubComponent(_args.predictorType);
trainer = temp.CreateInstance(Host) as TScalarTrainer;
}
_trainer = null;
_predictors[i] = TrainPredictor(ch, trainer, data, count);
}
}
return(CreatePredictor());
}
private protected override void CheckLabel(RoleMappedData data)
{
Contracts.AssertValue(data);
// REVIEW: For floating point labels, this will make a pass over the data.
// Should we instead leverage the pass made by the LBFGS base class? Ideally, it wouldn't
// make a pass over the data...
data.CheckMultiClassLabel(out _numClasses);
// Initialize prior counts.
_prior = new Double[_numClasses];
// Try to get the label key values metedata.
var labelCol = data.Schema.Label.Value;
var labelMetadataType = labelCol.Metadata.Schema.GetColumnOrNull(MetadataUtils.Kinds.KeyValues)?.Type;
if (!(labelMetadataType is VectorType vecType && vecType.ItemType == TextType.Instance && vecType.Size == _numClasses))
{
_labelNames = null;
return;
}
VBuffer <ReadOnlyMemory <char> > labelNames = default;
labelCol.Metadata.GetValue(MetadataUtils.Kinds.KeyValues, ref labelNames);
// If label names is not dense or contain NA or default value, then it follows that
// at least one class does not have a valid name for its label. If the label names we
// try to get from the metadata are not unique, we may also not use them in model summary.
// In both cases we set _labelNames to null and use the "Class_n", where n is the class number
// for model summary saving instead.
if (!labelNames.IsDense)
{
_labelNames = null;
return;
}
_labelNames = new string[_numClasses];
ReadOnlySpan <ReadOnlyMemory <char> > values = labelNames.GetValues();
// This hashset is used to verify the uniqueness of label names.
HashSet <string> labelNamesSet = new HashSet <string>();
for (int i = 0; i < _numClasses; i++)
{
ReadOnlyMemory <char> value = values[i];
if (value.IsEmpty)
{
_labelNames = null;
break;
}
var vs = values[i].ToString();
if (!labelNamesSet.Add(vs))
{
_labelNames = null;
break;
}
_labelNames[i] = vs;
Contracts.Assert(!string.IsNullOrEmpty(_labelNames[i]));
}
Contracts.Assert(_labelNames == null || _labelNames.Length == _numClasses);
}
private protected override void CheckLabel(RoleMappedData examples, out int weightSetCount)
{
examples.CheckMultiClassLabel(out weightSetCount);
}
protected override void CheckLabel(RoleMappedData examples)
{
examples.CheckMultiClassLabel(out _numClasses);
}
protected override void CheckLabel(RoleMappedData data)
{
Contracts.AssertValue(data);
// REVIEW: For floating point labels, this will make a pass over the data.
// Should we instead leverage the pass made by the LBFGS base class? Ideally, it wouldn't
// make a pass over the data...
data.CheckMultiClassLabel(out _numClasses);
// Initialize prior counts.
_prior = new Double[_numClasses];
// Try to get the label key values metedata.
var schema = data.Data.Schema;
var labelIdx = data.Schema.Label.Index;
var labelMetadataType = schema.GetMetadataTypeOrNull(MetadataUtils.Kinds.KeyValues, labelIdx);
if (labelMetadataType == null || !labelMetadataType.IsKnownSizeVector || !labelMetadataType.ItemType.IsText ||
labelMetadataType.VectorSize != _numClasses)
{
_labelNames = null;
return;
}
VBuffer <DvText> labelNames = default(VBuffer <DvText>);
schema.GetMetadata(MetadataUtils.Kinds.KeyValues, labelIdx, ref labelNames);
// If label names is not dense or contain NA or default value, then it follows that
// at least one class does not have a valid name for its label. If the label names we
// try to get from the metadata are not unique, we may also not use them in model summary.
// In both cases we set _labelNames to null and use the "Class_n", where n is the class number
// for model summary saving instead.
if (!labelNames.IsDense)
{
_labelNames = null;
return;
}
_labelNames = new string[_numClasses];
DvText[] values = labelNames.Values;
// This hashset is used to verify the uniqueness of label names.
HashSet <string> labelNamesSet = new HashSet <string>();
for (int i = 0; i < _numClasses; i++)
{
DvText value = values[i];
if (value.IsEmpty || value.IsNA)
{
_labelNames = null;
break;
}
var vs = values[i].ToString();
if (!labelNamesSet.Add(vs))
{
_labelNames = null;
break;
}
_labelNames[i] = vs;
Contracts.Assert(!string.IsNullOrEmpty(_labelNames[i]));
}
Contracts.Assert(_labelNames == null || _labelNames.Length == _numClasses);
}
protected override TVectorPredictor TrainPredictor(IChannel ch, TScalarTrainer trainer, RoleMappedData data, int count)
{
var data0 = data;
string dstName, labName;
var trans = MapLabelsAndInsertTransform(ch, data, out dstName, out labName, count, true, _args);
var newFeatures = trans.Schema.GetTempColumnName() + "NF";
var args3 = new DescribeTransform.Arguments {
columns = new string[] { labName, dstName }, oneRowPerColumn = true
};
var desc = new DescribeTransform(Host, args3, trans);
IDataView viewI;
if (_args.singleColumn && data.Schema.Label.Value.Type.RawKind() == DataKind.R4)
{
viewI = desc;
}
else if (_args.singleColumn)
{
var sch = new TypeReplacementSchema(desc.Schema, new[] { labName }, new[] { NumberType.R4 });
viewI = new TypeReplacementDataView(desc, sch);
#region debug
#if (DEBUG)
DebugChecking0(viewI, labName, false);
#endif
#endregion
}
else if (data.Schema.Label.Value.Type.IsKey())
{
int nb = data.Schema.Label.Value.Type.AsKey().KeyCount();
var sch = new TypeReplacementSchema(desc.Schema, new[] { labName }, new[] { new VectorType(NumberType.R4, nb) });
viewI = new TypeReplacementDataView(desc, sch);
#region debug
#if (DEBUG)
int nb_;
MinMaxLabelOverDataSet(trans, labName, out nb_);
int count3;
data.CheckMultiClassLabel(out count3);
if (count3 != nb)
{
throw ch.Except("Count mismatch (KeyCount){0} != {1}", nb, count3);
}
DebugChecking0(viewI, labName, true);
DebugChecking0Vfloat(viewI, labName, nb);
#endif
#endregion
}
else
{
int nb;
if (count <= 0)
{
MinMaxLabelOverDataSet(trans, labName, out nb);
}
else
{
nb = count;
}
var sch = new TypeReplacementSchema(desc.Schema, new[] { labName }, new[] { new VectorType(NumberType.R4, nb) });
viewI = new TypeReplacementDataView(desc, sch);
#region debug
#if (DEBUG)
DebugChecking0(viewI, labName, true);
#endif
#endregion
}
ch.Info("Merging column label '{0}' with features '{1}'", labName, data.Schema.Feature.Value.Name);
var args = string.Format("Concat{{col={0}:{1},{2}}}", newFeatures, data.Schema.Feature.Value.Name, labName);
IDataView after_concatenation = ComponentCreation.CreateTransform(Host, args, viewI);
var roles = data.Schema.GetColumnRoleNames()
.Where(kvp => kvp.Key.Value != RoleMappedSchema.ColumnRole.Label.Value)
.Where(kvp => kvp.Key.Value != RoleMappedSchema.ColumnRole.Feature.Value)
.Prepend(RoleMappedSchema.ColumnRole.Feature.Bind(newFeatures))
.Prepend(RoleMappedSchema.ColumnRole.Label.Bind(dstName));
var trainer_input = new RoleMappedData(after_concatenation, roles);
ch.Info("New Features: {0}:{1}", trainer_input.Schema.Feature.Value.Name, trainer_input.Schema.Feature.Value.Type);
ch.Info("New Label: {0}:{1}", trainer_input.Schema.Label.Value.Name, trainer_input.Schema.Label.Value.Type);
// We train the unique binary classifier.
var trainedPredictor = trainer.Train(trainer_input);
var predictors = new TScalarPredictor[] { trainedPredictor };
// We train the reclassification classifier.
if (_args.reclassicationPredictor != null)
{
var pred = CreateFinalPredictor(ch, data, trans, count, _args, predictors, null);
TrainReclassificationPredictor(data0, pred, ScikitSubComponent <ITrainer, SignatureTrainer> .AsSubComponent(_args.reclassicationPredictor));
}
return(CreateFinalPredictor(ch, data, trans, count, _args, predictors, _reclassPredictor));
}
protected override TVectorPredictor TrainPredictor(IChannel ch, TScalarTrainer trainer, RoleMappedData data, int count)
{
var data0 = data;
#region adding group ID
// We insert a group Id.
string groupColumnTemp = DataViewUtils.GetTempColumnName(data.Schema.Schema) + "GR";
var groupArgs = new GenerateNumberTransform.Arguments
{
Column = new[] { GenerateNumberTransform.Column.Parse(groupColumnTemp) },
UseCounter = true
};
var withGroup = new GenerateNumberTransform(Host, groupArgs, data.Data);
data = new RoleMappedData(withGroup, data.Schema.GetColumnRoleNames());
#endregion
#region preparing the training dataset
string dstName, labName;
var trans = MapLabelsAndInsertTransform(ch, data, out dstName, out labName, count, true, _args);
var newFeatures = trans.Schema.GetTempColumnName() + "NF";
// We check the label is not boolean.
int indexLab = SchemaHelper.GetColumnIndex(trans.Schema, dstName);
var typeLab = trans.Schema[indexLab].Type;
if (typeLab.RawKind() == DataKind.BL)
{
throw Host.Except("Column '{0}' has an unexpected type {1}.", dstName, typeLab.RawKind());
}
var args3 = new DescribeTransform.Arguments {
columns = new string[] { labName, dstName }, oneRowPerColumn = true
};
var desc = new DescribeTransform(Host, args3, trans);
IDataView viewI;
if (_args.singleColumn && data.Schema.Label.Value.Type.RawKind() == DataKind.R4)
{
viewI = desc;
}
else if (_args.singleColumn)
{
var sch = new TypeReplacementSchema(desc.Schema, new[] { labName }, new[] { NumberType.R4 });
viewI = new TypeReplacementDataView(desc, sch);
#region debug
#if (DEBUG)
DebugChecking0(viewI, labName, false);
#endif
#endregion
}
else if (data.Schema.Label.Value.Type.IsKey())
{
int nb = data.Schema.Label.Value.Type.AsKey().KeyCount();
var sch = new TypeReplacementSchema(desc.Schema, new[] { labName }, new[] { new VectorType(NumberType.R4, nb) });
viewI = new TypeReplacementDataView(desc, sch);
#region debug
#if (DEBUG)
int nb_;
MinMaxLabelOverDataSet(trans, labName, out nb_);
int count3;
data.CheckMultiClassLabel(out count3);
if (count3 != nb)
{
throw ch.Except("Count mismatch (KeyCount){0} != {1}", nb, count3);
}
DebugChecking0(viewI, labName, true);
DebugChecking0Vfloat(viewI, labName, nb);
#endif
#endregion
}
else
{
int nb;
if (count <= 0)
{
MinMaxLabelOverDataSet(trans, labName, out nb);
}
else
{
nb = count;
}
var sch = new TypeReplacementSchema(desc.Schema, new[] { labName }, new[] { new VectorType(NumberType.R4, nb) });
viewI = new TypeReplacementDataView(desc, sch);
#region debug
#if (DEBUG)
DebugChecking0(viewI, labName, true);
#endif
#endregion
}
ch.Info("Merging column label '{0}' with features '{1}'", labName, data.Schema.Feature.Value.Name);
var args = string.Format("Concat{{col={0}:{1},{2}}}", newFeatures, data.Schema.Feature.Value.Name, labName);
var after_concatenation_ = ComponentCreation.CreateTransform(Host, args, viewI);
#endregion
#region converting label and group into keys
// We need to convert the label into a Key.
var convArgs = new MultiClassConvertTransform.Arguments
{
column = new[] { MultiClassConvertTransform.Column.Parse(string.Format("{0}k:{0}", dstName)) },
keyRange = new KeyRange()
{
Min = 0, Max = 4
},
resultType = DataKind.U4
};
IDataView after_concatenation_key_label = new MultiClassConvertTransform(Host, convArgs, after_concatenation_);
// The group must be a key too!
convArgs = new MultiClassConvertTransform.Arguments
{
column = new[] { MultiClassConvertTransform.Column.Parse(string.Format("{0}k:{0}", groupColumnTemp)) },
keyRange = new KeyRange()
{
Min = 0, Max = null
},
resultType = _args.groupIsU4 ? DataKind.U4 : DataKind.U8
};
after_concatenation_key_label = new MultiClassConvertTransform(Host, convArgs, after_concatenation_key_label);
#endregion
#region preparing the RoleMapData view
string groupColumn = groupColumnTemp + "k";
dstName += "k";
var roles = data.Schema.GetColumnRoleNames();
var rolesArray = roles.ToArray();
roles = roles
.Where(kvp => kvp.Key.Value != RoleMappedSchema.ColumnRole.Label.Value)
.Where(kvp => kvp.Key.Value != RoleMappedSchema.ColumnRole.Feature.Value)
.Where(kvp => kvp.Key.Value != groupColumn)
.Where(kvp => kvp.Key.Value != groupColumnTemp);
rolesArray = roles.ToArray();
if (rolesArray.Any() && rolesArray[0].Value == groupColumnTemp)
{
throw ch.Except("Duplicated group.");
}
roles = roles
.Prepend(RoleMappedSchema.ColumnRole.Feature.Bind(newFeatures))
.Prepend(RoleMappedSchema.ColumnRole.Label.Bind(dstName))
.Prepend(RoleMappedSchema.ColumnRole.Group.Bind(groupColumn));
var trainer_input = new RoleMappedData(after_concatenation_key_label, roles);
#endregion
ch.Info("New Features: {0}:{1}", trainer_input.Schema.Feature.Value.Name, trainer_input.Schema.Feature.Value.Type);
ch.Info("New Label: {0}:{1}", trainer_input.Schema.Label.Value.Name, trainer_input.Schema.Label.Value.Type);
// We train the unique binary classifier.
var trainedPredictor = trainer.Train(trainer_input);
var predictors = new TScalarPredictor[] { trainedPredictor };
// We train the reclassification classifier.
if (_args.reclassicationPredictor != null)
{
var pred = CreateFinalPredictor(ch, data, trans, count, _args, predictors, null);
TrainReclassificationPredictor(data0, pred, ScikitSubComponent <ITrainer, SignatureTrainer> .AsSubComponent(_args.reclassicationPredictor));
}
return(CreateFinalPredictor(ch, data, trans, count, _args, predictors, _reclassPredictor));
}
