book

Time Series Forecasting in Python

by Marco Peixeiro

October 2022

Beginner to intermediate

456 pages

12h 12m

English

Manning Publications

Read now

Unlock full access

prefaceacknowledgmentsabout this bookWho should read this book?How this book is organized: A roadmapAbout the codeliveBook discussion forumAuthor onlineabout the authorabout the cover illustration
1.1 Introducing time series1.1.1 Components of a time series1.2 Bird’s-eye view of time series forecasting1.2.1 Setting a goal1.2.2 Determining what must be forecast to achieve your goal1.2.3 Setting the horizon of the forecast1.2.4 Gathering the data1.2.5 Developing a forecasting model1.2.6 Deploying to production1.2.7 Monitoring1.2.8 Collecting new data1.3 How time series forecasting is different from other regression tasks1.3.1 Time series have an order1.3.2 Time series sometimes do not have features1.4 Next stepsSummary
2.1 Defining a baseline model2.2 Forecasting the historical mean2.2.1 Setup for baseline implementations2.2.2 Implementing the historical mean baseline2.3 Forecasting last year’s mean2.4 Predicting using the last known value2.5 Implementing the naive seasonal forecast2.6 Next stepsSummary
3.1 The random walk process3.1.1 Simulating a random walk process3.2 Identifying a random walk3.2.1 Stationarity3.2.2 Testing for stationarity3.2.3 The autocorrelation function3.2.4 Putting it all together3.2.5 Is GOOGL a random walk?3.3 Forecasting a random walk3.3.1 Forecasting on a long horizon3.3.2 Forecasting the next timestep3.4 Next steps3.5 Exercises3.5.1 Simulate and forecast a random walk3.5.2 Forecast the daily closing price of GOOGL3.5.3 Forecast the daily closing price of a stock of your choiceSummary

4.1 Defining a moving average process4.1.1 Identifying the order of a moving average process4.2 Forecasting a moving average process4.3 Next steps4.4 Exercises4.4.1 Simulate an MA(2) process and make forecasts4.4.2 Simulate an MA(q) process and make forecastsSummary
5.1 Predicting the average weekly foot traffic in a retail store5.2 Defining the autoregressive process5.3 Finding the order of a stationary autoregressive process5.3.1 The partial autocorrelation function (PACF)5.4 Forecasting an autoregressive process5.5 Next steps5.6 Exercises5.6.1 Simulate an AR(2) process and make forecasts5.6.2 Simulate an AR(p) process and make forecastsSummary
6.1 Forecasting bandwidth usage for data centers6.2 Examining the autoregressive moving average process6.3 Identifying a stationary ARMA process6.4 Devising a general modeling procedure6.4.1 Understanding the Akaike information criterion (AIC)6.4.2 Selecting a model using the AIC6.4.3 Understanding residual analysis6.4.4 Performing residual analysis6.5 Applying the general modeling procedure6.6 Forecasting bandwidth usage6.7 Next steps6.8 Exercises6.8.1 Make predictions on the simulated ARMA(1,1) process6.8.2 Simulate an ARMA(2,2) process and make forecastsSummary
7.1 Defining the autoregressive integrated moving average model7.2 Modifying the general modeling procedure to account for non-stationary series7.3 Forecasting a non-stationary times series7.4 Next steps7.5 Exercises7.5.1 Apply the ARIMA(p,d,q) model on the datasets from chapters 4, 5, and 6Summary
8.1 Examining the SARIMA(p,d,q)(P,D,Q)m model8.2 Identifying seasonal patterns in a time series8.3 Forecasting the number of monthly air passengers8.3.1 Forecasting with an ARIMA(p,d,q) model8.3.2 Forecasting with a SARIMA(p,d,q)(P,D,Q)m model8.3.3 Comparing the performance of each forecasting method8.4 Next steps8.5 Exercises8.5.1 Apply the SARIMA(p,d,q)(P,D,Q)m model on the Johnson & Johnson datasetSummary
9.1 Examining the SARIMAX model9.1.1 Exploring the exogenous variables of the US macroeconomics dataset9.1.2 Caveat for using SARIMAX9.2 Forecasting the real GDP using the SARIMAX model9.3 Next steps9.4 Exercises9.4.1 Use all exogenous variables in a SARIMAX model to predict the real GDPSummary
10.1 Examining the VAR model10.2 Designing a modeling procedure for the VAR(p) model10.2.1 Exploring the Granger causality test10.3 Forecasting real disposable income and real consumption10.4 Next steps10.5 Exercises10.5.1 Use a VARMA model to predict realdpi and realcons10.5.2 Use a VARMAX model to predict realdpi and realconsSummary
11.1 Importing the required libraries and loading the data11.2 Visualizing the series and its components11.3 Modeling the data11.3.1 Performing model selection11.3.2 Conducting residual analysis11.4 Forecasting and evaluating the model’s performanceNext steps
12.1 When to use deep learning for time series forecasting12.2 Exploring the different types of deep learning models12.3 Getting ready to apply deep learning for forecasting12.3.1 Performing data exploration12.3.2 Feature engineering and data splitting12.4 Next steps12.5 ExerciseSummary
13.1 Creating windows of data13.1.1 Exploring how deep learning models are trained for time series forecasting13.1.2 Implementing the DataWindow class13.2 Applying baseline models13.2.1 Single-step baseline model13.2.2 Multi-step baseline models13.2.3 Multi-output baseline model13.3 Next steps13.4 ExercisesSummary
14.1 Implementing a linear model14.1.1 Implementing a single-step linear model14.1.2 Implementing a multi-step linear model14.1.3 Implementing a multi-output linear model14.2 Implementing a deep neural network14.2.1 Implementing a deep neural network as a single-step model14.2.2 Implementing a deep neural network as a multi-step model14.2.3 Implementing a deep neural network as a multi-output model14.3 Next steps14.4 ExercisesSummary
15.1 Exploring the recurrent neural network (RNN)15.2 Examining the LSTM architecture15.2.1 The forget gate15.2.2 The input gate15.2.3 The output gate15.3 Implementing the LSTM architecture15.3.1 Implementing an LSTM as a single-step model15.3.2 Implementing an LSTM as a multi-step model15.3.3 Implementing an LSTM as a multi-output model15.4 Next steps15.5 ExercisesSummary
16.1 Examining the convolutional neural network (CNN)16.2 Implementing a CNN16.2.1 Implementing a CNN as a single-step model16.2.2 Implementing a CNN as a multi-step model16.2.3 Implementing a CNN as a multi-output model16.3 Next steps16.4 ExercisesSummary
17.1 Examining the ARLSTM architecture17.2 Building an autoregressive LSTM model17.3 Next steps17.4 ExercisesSummary
18.1 Understanding the capstone project18.1.1 Objective of this capstone project18.2 Data wrangling and preprocessing18.2.1 Dealing with missing data18.2.2 Data conversion18.2.3 Data resampling18.3 Feature engineering18.3.1 Removing unnecessary columns18.3.2 Identifying the seasonal period18.3.3 Splitting and scaling the data18.4 Preparing for modeling with deep learning18.4.1 Initial setup18.4.2 Defining the DataWindow class18.4.3 Utility function to train our models18.5 Modeling with deep learning18.5.1 Baseline models18.5.2 Linear model18.5.3 Deep neural network18.5.4 Long short-term memory (LSTM) model18.5.5 Convolutional neural network (CNN)18.5.6 Combining a CNN with an LSTM18.5.7 The autoregressive LSTM model18.5.8 Selecting the best model18.6 Next steps
19.1 Overview of the automated forecasting libraries19.2 Exploring Prophet19.3 Basic forecasting with Prophet19.4 Exploring Prophet’s advanced functionality19.4.1 Visualization capabilities19.4.2 Cross-validation and performance metrics19.4.3 Hyperparameter tuning19.5 Implementing a robust forecasting process with Prophet19.5.1 Forecasting project: Predicting the popularity of “chocolate” searches on Google19.5.2 Experiment: Can SARIMA do better?19.6 Next steps19.7 Exercises19.7.1 Forecast the number of air passengers19.7.2 Forecast the volume of antidiabetic drug prescriptions19.7.3 Forecast the popularity of a keyword on Google TrendsSummary
20.1 Understanding the capstone project20.1.1 Objective of the capstone project20.2 Data preprocessing and visualization20.3 Modeling with Prophet20.4 Optional: Develop a SARIMA model20.5 Next steps
21.1 Summarizing what you’ve learned21.1.1 Statistical methods for forecasting21.1.2 Deep learning methods for forecasting21.1.3 Automating the forecasting process21.2 What if forecasting does not work?21.3 Other applications of time series data21.4 Keep practicing
Installing AnacondaPythonJupyter NotebooksGitHub RepositoryInstalling ProphetInstalling libraries in Anaconda

Content preview from Time Series Forecasting in Python

4 Modeling a moving average process

This chapter covers

Defining a moving average process
Using the ACF to identify the order of a moving average process
Forecasting a time series using the moving average model

In the previous chapter, you learned how to identify and forecast a random walk process. We defined a random walk process as a series whose first difference is stationary with no autocorrelation. This means that plotting its ACF will show no significant coefficients after lag 0. However, it is possible that a stationary process may still exhibit autocorrelation. In this case, we have a time series that can be approximated by a moving average model MA(q)), an autoregressive model AR(p)), or an autoregressive moving average model ARMA( ...