Main Content

增量学习

将分类模型与流化数据进行拟合并跟踪其性能

增量学习（或在线学习）需要处理来自数据流的传入数据，且可能对预测变量的分布、目标函数的各个方面，以及观测值是否带标签等知识都知之甚少或完全不了解。增量学习问题与传统的机器学习方法形成对比，在传统的机器学习方法中，有足够的带标签数据可用于模型拟合、执行交叉验证以调整超参数并推断预测变量分布特性。

增量学习需要经过配置的增量模型。您可以通过调用一个对象（例如 incrementalClassificationLinear）直接创建和配置增量模型，也可以使用 incrementalLearner 将受支持的以传统方式训练的模型转换为增量学习器。在配置模型并设置数据流后，您可以将增量模型与传入的数据块进行拟合，跟踪该模型的预测性能，或同时执行这两项操作。

有关详细信息，请参阅Incremental Learning Overview。

您还能够以增量方式监控概念数据中的漂移，例如分类误差。首先，您需要使用 incrementalConceptDriftDetector 配置漂移检测器。在建立数据流后，您可以更新漂移检测器并使用 detectdrift 检查任何漂移。有关详细信息，请参阅参考页。

模块

IncrementalClassificationLinear PredictClassify observations using incremental linear classification model (自 R2023b 起)
IncrementalClassificationLinear FitFit incremental linear binary classification model (自 R2023b 起)
IncrementalClassificationECOC PredictClassify observations using incremental ECOC classification model (自 R2024a 起)
IncrementalClassificationECOC FitFit incremental ECOC classification model (自 R2024a 起)
Update MetricsUpdate performance metrics in incremental learning model given new data (自 R2023b 起)

函数

全部展开

创建增量漂移感知模型

incrementalDriftAwareLearnerConstruct drift-aware model for incremental learning (自 R2022b 起)

以增量方式拟合和跟踪性能

fitTrain drift-aware learner for incremental learning with new data (自 R2022b 起)
updateMetricsUpdate performance metrics in incremental drift-aware learning model given new data (自 R2022b 起)
updateMetricsAndFitUpdate performance metrics in incremental drift-aware learning model given new data and train model (自 R2022b 起)

其他模型操作

lossRegression or classification error of incremental drift-aware learner (自 R2022b 起)
perObservationLossPer observation regression or classification error of incremental drift-aware learner (自 R2022b 起)
predictPredict responses for new observations from incremental drift-aware learning model (自 R2022b 起)
resetReset incremental drift-aware learner (自 R2022b 起)

创建增量模型

incrementalClassificationKernel Binary classification kernel model for incremental learning (自 R2022a 起)
incrementalLearnerConvert kernel model for binary classification to incremental learner (自 R2022a 起)

以增量方式拟合和跟踪性能

fitTrain kernel model for incremental learning (自 R2022a 起)
updateMetricsUpdate performance metrics in kernel incremental learning model given new data (自 R2022a 起)
updateMetricsAndFitUpdate performance metrics in kernel incremental learning model given new data and train model (自 R2022a 起)

其他模型操作

predictPredict responses for new observations from kernel incremental learning model (自 R2022a 起)
lossLoss of kernel incremental learning model on batch of data (自 R2022a 起)
perObservationLossPer observation classification error of model for incremental learning (自 R2022a 起)
resetReset incremental classification model (自 R2022a 起)

创建增量模型

incrementalClassificationLinearBinary classification linear model for incremental learning (自 R2020b 起)
incrementalLearnerConvert binary classification support vector machine (SVM) model to incremental learner (自 R2020b 起)
incrementalLearnerConvert linear model for binary classification to incremental learner (自 R2020b 起)

以增量方式拟合和跟踪性能

fitTrain linear model for incremental learning (自 R2020b 起)
updateMetricsUpdate performance metrics in linear incremental learning model given new data (自 R2020b 起)
updateMetricsAndFitUpdate performance metrics in linear incremental learning model given new data and train model (自 R2020b 起)

其他模型操作

predictPredict responses for new observations from linear incremental learning model (自 R2020b 起)
lossLoss of linear incremental learning model on batch of data (自 R2020b 起)
perObservationLossPer observation classification error of model for incremental learning (自 R2022a 起)
resetReset incremental classification model (自 R2022a 起)

创建增量模型

incrementalClassificationECOC Multiclass classification model using binary learners for incremental learning (自 R2022a 起)
incrementalLearnerConvert multiclass error-correcting output codes (ECOC) model to incremental learner (自 R2022a 起)

以增量方式拟合和跟踪性能

fitTrain ECOC classification model for incremental learning (自 R2022a 起)
updateMetricsUpdate performance metrics in ECOC incremental learning classification model given new data (自 R2022a 起)
updateMetricsAndFitUpdate performance metrics in ECOC incremental learning classification model given new data and train model (自 R2022a 起)

其他模型操作

predictPredict responses for new observations from ECOC incremental learning classification model (自 R2022a 起)
lossLoss of ECOC incremental learning classification model on batch of data (自 R2022a 起)
perObservationLossPer observation classification error of model for incremental learning (自 R2022a 起)
resetReset incremental classification model (自 R2022a 起)

创建增量模型

incrementalClassificationNaiveBayesNaive Bayes classification model for incremental learning (自 R2021a 起)
incrementalLearnerConvert naive Bayes classification model to incremental learner (自 R2021a 起)

以增量方式拟合和跟踪性能

fitTrain naive Bayes classification model for incremental learning (自 R2021a 起)
updateMetricsUpdate performance metrics in naive Bayes incremental learning classification model given new data (自 R2021a 起)
updateMetricsAndFitUpdate performance metrics in naive Bayes incremental learning classification model given new data and train model (自 R2021a 起)

其他模型操作

predictPredict responses for new observations from naive Bayes incremental learning classification model (自 R2021a 起)
lossLoss of naive Bayes incremental learning classification model on batch of data (自 R2021a 起)
logpLog unconditional probability density of naive Bayes classification model for incremental learning (自 R2021a 起)
perObservationLossPer observation classification error of model for incremental learning (自 R2022a 起)
resetReset incremental classification model (自 R2022a 起)

创建概念漂移检测器

incrementalConceptDriftDetectorInstantiate incremental concept drift detector (自 R2022a 起)

检测漂移并重置模型

detectdriftUpdate drift detector states and drift status with new data (自 R2022a 起)
resetReset incremental concept drift detector (自 R2022a 起)

对象

全部展开

DriftDetectionMethodIncremental drift detector that utilizes Drift Detection Method (DDM) (自 R2022a 起)
HoeffdingDriftDetectionMethodIncremental concept drift detector that utilizes Hoeffding's Bounds Drift Detection Method (HDDM) (自 R2022a 起)

主题