// 从options中创建训练器
func NewOnlineSGDClassifier(options OnlineSGDClassifierOptions) *OnlineSGDClassifier {
classifier := new(OnlineSGDClassifier)
classifier.options = options
if classifier.options.BatchSize <= 1 {
classifier.options.BatchSize = 1
}
classifier.weights = util.NewSparseMatrix(options.NumLabels - 1)
classifier.derivative = util.NewSparseMatrix(options.NumLabels - 1)
classifier.instanceDerivative = util.NewSparseMatrix(options.NumLabels - 1)
classifier.evaluator = new(FrapEvaluator)
classifier.evaluator.Init(options.NumInstancesForEvaluation)
classifier.featureDictionary = dictionary.NewDictionary(1)
classifier.labelDictionary = dictionary.NewDictionary(0)
return classifier
}
// 向数据集中添加一个样本
// 成功添加则返回true，否则返回false
func (set *inmemDataset) AddInstance(instance *Instance) bool {
set.CheckFinalized(false)
// 添加第一条样本时确定数据集的一些性质
if len(set.instances) == 0 {
if instance.NamedFeatures != nil {
set.useFeatureDict = true
set.featureDict = dictionary.NewDictionary(1) // 特征ID从0开始
ConvertNamedFeatures(instance, set.featureDict)
}
if instance.Features.IsSparse() {
set.options.FeatureIsSparse = true
set.options.FeatureDimension = 0
} else {
set.options.FeatureIsSparse = false
set.options.FeatureDimension = len(instance.Features.Keys())
}
if instance.Output == nil {
set.options.IsSupervisedLearning = false
} else {
set.options.IsSupervisedLearning = true
if instance.Output.LabelString != "" {
set.useLabelDict = true
set.labelDict = dictionary.NewDictionary(0)
instance.Output.Label =
set.labelDict.GetIdFromName(instance.Output.LabelString)
}
}
} else {
// 否则检查后续数据样本类型是否一致
if instance.NamedFeatures != nil {
ConvertNamedFeatures(instance, set.featureDict)
if !set.useFeatureDict {
log.Print("数据集不使用特征词典而添加的样本使用NamedFeatures")
return false
}
} else {
if set.useFeatureDict {
log.Print("数据集使用特征词典而添加的样本不使用NamedFeatures")
return false
}
}
if set.options.FeatureIsSparse {
if !instance.Features.IsSparse() {
log.Print("数据集使用稀疏特征而添加的样本不稀疏")
return false
}
} else {
if instance.Features.IsSparse() {
log.Print("数据集使用稠密特征而添加的样本稀疏")
return false
}
if set.options.FeatureDimension != len(instance.Features.Keys()) {
log.Print("数据集特征数和添加样本的特征数不同")
return false
}
}
if instance.Output == nil {
if set.options.IsSupervisedLearning {
log.Print("数据集为监督式而添加样本为非监督式数据")
return false
}
} else {
if !set.options.IsSupervisedLearning {
log.Print("数据集为非监督式而添加样本为监督式数据")
return false
}
if instance.Output.LabelString != "" {
if !set.useLabelDict {
log.Print("数据集不使用标注词典而添加的样本使用LabelString")
return false
}
} else {
if set.useLabelDict {
log.Print("数据集使用标注词典而添加的样本不使用LabelString")
return false
}
}
}
}
if set.options.IsSupervisedLearning {
if instance.Output.LabelString != "" {
instance.Output.Label =
set.labelDict.GetIdFromName(instance.Output.LabelString)
}
if instance.Output.Label < 0 {
log.Println("样本标注值不在合法范围")
return false
}
if instance.Output.Label >= set.options.NumLabels {
set.options.NumLabels = instance.Output.Label + 1
}
}
set.instances = append(set.instances, instance)
return true
}