book

Machine Learning for Time Series with Python - Second Edition

Name: Machine Learning for Time Series with Python - Second Edition
Author: Dr. Ben Auffarth
ISBN: 9781837631339

by Dr. Ben Auffarth

October 2026

Intermediate to advanced

486 pages

12h 21m

English

Packt Publishing

Read now

Unlock full access

Preface
Who this book is forWhat this book coversTo get the most out of this bookGet in touchFree benefits with your bookHow to unlock
Towards Modern Forecasting
The evolution of forecasting methodsWhat is a time series?Time series in the Python contextWhen do smart models fail?Why algorithms aren’t enoughWhy advanced models still failFailure mode 1: misleading validation metricsFailure mode 2: integration breakdownFailure mode 3: inability to adapt to changeThe three fundamental gapsGap 1: temporal assumption violationsGap 2: misalignment with business needsGap 3: the problem with fixed modelsBuilding adaptive systemsQuestion 1: is this problem actually foreseeable?Question 2: what decisions will this prediction enable?Question 3: how will patterns change over time?Question 4: how quickly can I adapt when I’m wrong?Why choose Python?Choosing the right tools for time series problemsForecastingClassificationAnomaly detectionRegressionCore principles for production-ready systemsCommon pitfalls and how to avoid themPattern 1: static system thinking to dynamic system designPattern 2: optimization mindset to adaptation infrastructurePattern 3: point estimate fixation to uncertainty-aware decision-makingPattern 4: independence assumptions to temporal structure respectPattern 5: technical metric obsession to business value alignmentPattern 6: model artifact focus to production system designSummaryFurther ReadingGet this book’s PDF version and more
Preparing and Visualizing Time Series Data
Technical requirementsMathematical foundations for data understandingTemporal dependencies and autocorrelationProving temporal structure existsInterpreting autocorrelation patterns for modeling strategiesTools for autocorrelation analysisStationarity testing and interpretationVisualizing stationarityStatistical stationarity testsAchieving stationarityStabilizing variance with log and Box–Cox transformationsSignals from decompositionWhy the additive vs. multiplicative choice mattersPractical decomposition in PythonChoosing the right decomposition methodBuilding a time series data pipelineTemporal indexing and frequency managementThe strategic solution: robust conversion and validationQuality checks and missing dataMissing data assessmentDetecting structural breaksQuality threshold decision-makingFrequency conversion and resampling decisionsThe resampling trap: Look-Ahead biasPractical decision frameworkPattern discovery through visualizationVisual diagnostics for temporal structuresAutocorrelation and residual analysisForecastability assessmentMeasuring predictability before modelingSignal-to-noise proxyThe naive modelThe seasonal naive benchmarkA framework for project viabilitySummary
Classical Models and Validation
Why statistical methods matterThe role of baselines in forecastingClassical baselinesExponential smoothingARIMATheta MethodOverview of statistical forecasting methodsCase study: forecasting “remote work”Validation and metricsWhy random splits failExpanding windowRolling windowAdvanced techniquesWalk-forward validationNested cross-validationCommon pitfallsPerformance metricsScale-dependent metricsScale-independent metricsAn experimentIndustry-specific validation strategiesSummaryGet this book’s PDF version and more
Forecasting with Machine Learning
Forecasting as a regression taskThe ML forecasting workflowStep 1: data preparationStep 2: feature engineeringStep 3: model trainingStep 4: predictionStep 5: validationThe role of toolingTools for the jobSktimeDartsThe Nixtla ecosystemAutoML platforms Direct scikit-learn integration Choosing your tool: a decision frameworkCreating an ML forecastStep 1: setup and baselineStep 2: manual feature engineeringStep 3: training the modelStep 4: predictionStep 5: validationSummary
Feature Engineering and Tree-Based Models
Technical requirementsFeature engineeringCalendar featuresEvent and holiday featuresAdvanced feature engineeringLag featuresRolling window statisticsEncoding cyclical timeDomain-specific featuresA practical guide to feature engineeringBuilding a reusable pipelineChoosing a tree modelHybrid modeling and decompositionPattern 1: Decompose-then-model-residualsPattern 2: Fourier features for seasonalityAutomated feature engineeringRecipe 1: mlforecast and XGBoostRecipe 2: tsfresh and DartsRecipe 3: functimeSummaryGet this book’s PDF version and more
Multivariate and Hierarchical Forecasting
Multivariate modelsWhat are multivariate forecasts?VAR and VARMAXA simple VAR modelGranger causalityVARMAXGlobal multi-series modelsHierarchical modelsWhat are hierarchies?Reconciliation approachesThe bottom-up approachThe top-down approachThe middle-out approachModern reconciliation: optimal combinationsCase study: Forecasting tourism in AustraliaEvaluating and comparing resultsSummaryFurther Reading
Practical Deep Learning for Time Series
Technical requirementsWhy deep learning?Foundational modelsLong short-term memoryTemporal Convolutional NetworksLinear and MLP-based modelsDLinearN-BEATSTSMixerTransformer modelsTemporal fusion transformer (TFT)PatchTSTiTransformerCrossformer2D and decomposition modelsModel selectionPractical tips for training deep learning modelsFurther ReadingSummaryGet this book’s PDF version and more
Quantifying Time Series Uncertainty with Conformal Prediction
The problem with uncertaintyThe conformal prediction frameworkThe conformal guaranteeThe split conformal algorithmFixed-width intervals with mapieAdaptive intervals with conformalized quantile regression (CQR)The failure of fixed-width intervalsImplementation with mapieRecipes for the modern time series ecosystemRecipe 1: nixtla/statsforecastRecipe 2: sktimeRecipe 3: DARTS (native conformal models)Conformal prediction for non-exchangeable time seriesEnbPI in practiceACI in practiceMulti-horizon coverageProduction-ready uncertaintyEvaluating interval qualityMonitoring for calibration driftBuilding a rolling coverage dashboardRecalibration strategyDiagnosing conditional coverageModeling financial volatilityIdentifying volatility clusteringA hybrid model: GARCH combined with conformal predictionValidating exchangeability of standardized residualsProduction and deployment patternsChoosing your uncertainty methodSummaryFurther Reading
Foundation Models: Quantitative and Qualitative Forecasting
A taxonomy of foundation model architecturesThe need for time series foundation modelsA hands-on exampleQuantitative foundation modelsGoogle TimesFMAmazon ChronosSalesforce MoiraiQualitative and multimodal forecastingIntroducing OpenTSLMPrompt engineering for multimodal LLMs: Chain-of-ChartsMultimodal datasetsCritical challengesPerformance risksGeneralization risksMemorization risksAdversarial fragilityProbabilistic calibrationLong-horizon decayDeploying foundation models in productionSummaryGet this book’s PDF version and more

Production Workflows: Deployment, Monitoring, and Scaling
The production gapArchitectural patterns for time-series deploymentBatch processing architecturesThe schedule-compute-cache patternGlobal context and robustnessStream processing (real-time) architectureEvent-driven inference and state managementThe training-serving skew challengeHybrid approaches: lambda and kappa architecturesData engineeringPoint-in-time correctness and time travelThe Feature Store solutionHandling data gaps, irregularities, and censored dataData validation with PanderaScaling strategiesLocal vs. global modeling paradigmsDistributed forecasting with rayHierarchical reconciliation at scaleDeployment strategiesShadow deploymentModel serialization and MLflowMonitoring and observabilityProxy metrics and drift detectionPerformance estimation (CBPE)Advanced failure modes and mitigationCensored data and the demand death spiralFeedback loops and the bullwhip effectSummary
Beyond Forecasting: Specialized Applications
Time series classification and structural pattern recognitionSelf-supervised representationsktime vs. aeonThe human activity recognition problemDiffusion models for time seriesManaging nonstationarityDrift detection methodsImplementation: adaptive windowing (ADWIN)Graph and spatiotemporal modelingNeural ODEs for irregular time seriesCausal inferenceIntervention analysis with CausalImpactBenchmarksSummaryGet this book’s PDF version and more
Intermittent Forecasting and Survival Analysis
Intermittent demand forecastingThe mathematics of sparsity and obsolescenceHierarchical Bayesian formulationsImplementation of intermittent heuristicsStacking and hurdle-based combinationsDeep learning architectures for zero-inflated dataTime-to-event and survival analysisThe challenge of censoring and truncationMathematical foundations: hazard and state-space transformationsImplementation of the Cox Proportional Hazards modelMulti-horizon forecasting and tree-based survivalEvaluating time-to-event calibrationDiscrete-time hazard forecasting at scaleApplied survival and reconciliationSummary
Unlock Your Exclusive Benefits
Unlock this Book’s Free Benefits in 3 Easy Steps
Other Books You May Enjoy
Index

Content preview from Machine Learning for Time Series with Python - Second Edition

10 Production Workflows: Deployment, Monitoring, and Scaling

We have spent the previous chapters exploring the wide range of modeling options, from the statistical rigor of ARIMA in Chapter 3 to the zero-shot capabilities of Foundation Models in Chapter 9. In those contexts, success was defined by a single metric: accuracy on a hold-out test set.

However, a model residing in a Jupyter notebook is not a product. The transition from a research environment as defined by static CSV files and clean validation sets to a production environment defined by unbounded data streams, strict latency SLAs, and shifting distributions is often where time-series projects fail.

In the industry, the practice of developing and deploying models is often separated ...

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Machine Learning for Time-Series with Python

Publisher Resources

ISBN: 9781837631339

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills

Machine Learning for Time Series with Python - Second Edition

by Dr. Ben Auffarth

10

Production Workflows: Deployment, Monitoring, and Scaling

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.