book

Machine Learning for Time-Series with Python

by Ben Auffarth

October 2021

Beginner to intermediate

370 pages

8h 19m

English

Packt Publishing

Read now

Unlock full access

Who this book is forWhat this book coversTo get the most out of this bookGet in touch
What Is a Time-Series?Characteristics of Time-SeriesTime-Series and Forecasting – Past and PresentDemographyGeneticsAstronomyEconomicsMeteorologyMedicineApplied StatisticsPython for Time-SeriesInstalling librariesJupyter Notebook and JupyterLabNumPypandasBest practice in PythonSummary
What is time-series analysis?Working with time-series in PythonRequirementsDatetimepandasUnderstanding the variablesUncovering relationships between variablesIdentifying trend and seasonalitySummary
What Is Preprocessing?Feature TransformsScalingLog and Power TransformationsImputationFeature EngineeringDate- and Time-Related FeaturesROCKETShapeletsPython PracticeLog and Power Transformations in PracticeImputationHoliday FeaturesDate AnnotationPaydaysSeasonsThe Sun and MoonBusiness DaysAutomated Feature ExtractionROCKETShapelets in PracticeSummary
Machine learning with time-seriesSupervised, unsupervised, and reinforcement learning History of machine learningMachine learning workflowCross-validation Error metrics for time-seriesRegressionClassificationComparing time-seriesMachine learning algorithms for time-seriesDistance-based approachesShapeletsROCKETTime-Series Forest and Canonical Interval ForestSymbolic approachesHIVE-COTEDiscussionImplementationsSummary
What are classical models?Moving average and autoregressionModel selection and orderExponential smoothingARCH and GARCHVector autoregressionPython librariesStatsmodelsPython practiceRequirementsModeling in PythonSummary
Unsupervised methods for time-seriesAnomaly detectionMicrosoftGoogleAmazonFacebookTwitterImplementationsChange point detectionClusteringPython practiceRequirementsAnomaly detectionChange point detectionSummary
More machine learning methods for time-seriesValidationK-nearest neighbors with dynamic time warpingSilverkiteGradient boostingPython exerciseVirtual environmentsK-nearest neighbors with dynamic time warping in PythonSilverkiteGradient boostingEnsembles with KatsSummary
Online learning for time-seriesOnline algorithmsDriftDrift detection methodsAdaptive learning methodsPython practiceDrift detectionRegressionModel selectionSummary
Probabilistic Models for Time-SeriesProphetMarkov ModelsFuzzy ModelingBayesian Structural Time-Series ModelsPython ExerciseProphetMarkov Switching ModelFuzzy Time-SeriesBayesian Structural Time-Series ModelingSummary

Introduction to deep learningDeep learning for time-seriesAutoencodersInceptionTimeDeepARN-BEATSRecurrent neural networksConvNetsTransformer architecturesInformerPython practiceFully connected networkRecurrent neural networkDilated causal convolutional neural networkSummary
Introduction to reinforcement learningReinforcement Learning for Time-SeriesBandit algorithmsDeep Q-LearningPython PracticeRecommendationsTrading with DQNSummary
Forecasting a Multivariate Time-SeriesPython practiceWhat's next for time-series?

Content preview from Machine Learning for Time-Series with Python

8 Online Learning for Time-Series

In this chapter, we are going to dive into online learning and streaming data for time-series. Online learning means that we continually update our model as new data is coming in. The advantage of online learning algorithms is that they can handle the high speed and possibly large size of streaming data and are able to adapt to new distributions of the data.

We will discuss drift, which is important because the performance of a machine learning model can be strongly affected by changes to the dataset to the point that a model will become obsolete (stale).

We are going to discuss what online learning is, how data can change (drift), and how adaptive learning algorithms combine drift detection methods to adjust ...