Deep Learning for Time Series Cookbook

Packt via Coursera

Go to class Write review

Overview

Google, IBM & Meta Certificates – 40% Off

One plan covers every Professional Certificate on Coursera.

Unlock All Certificates

Deep Learning for Time Series Cookbook is a hands-on course that helps you tackle a variety of time series problems using deep learning through practical coding recipes. You'll learn how to develop accurate forecasting models and extract insights from temporal data using the PyTorch ecosystem. Throughout this course, you'll explore essential concepts and architectures, including CNNs, transformers, autoencoders, and PyTorch Lightning, gaining practical experience in building models for forecasting, classification, and anomaly detection. The step-by-step recipes guide you from preprocessing time series data to creating production-ready predictive solutions. The course emphasizes real-world applications, showing how deep learning can uncover complex patterns, improve predictions, and optimize decision-making for univariate and multivariate datasets. Each module blends theory with hands-on exercises to reinforce understanding and ensure skills are immediately applicable. This course is ideal for machine learning enthusiasts, data scientists, and AI professionals looking to enhance their skills in time series analysis. A basic knowledge of Python and foundational machine learning concepts is recommended to get the most out of the content.

Syllabus

Getting Started with Time Series

This module introduces the foundational concepts and techniques of time series analysis, including data loading, visualization, decomposition, and handling non-stationary data. Learners will gain practical skills in using Python to explore, preprocess, and analyze time series datasets for meaningful insights.

Getting Started with PyTorch

This module introduces learners to the fundamentals of PyTorch, focusing on tensor operations, neural network construction, and training techniques. You will explore how to build and train feedforward, recurrent, and LSTM neural networks for tasks such as time series prediction. By the end, you'll gain hands-on experience with core deep learning workflows in PyTorch.

Univariate Time Series Forecasting

This module introduces key techniques for forecasting univariate time series, including classical ARIMA models and advanced deep learning approaches such as feedforward neural networks, LSTMs, and GRUs. Learners will gain practical skills in preparing time series data for supervised learning and handling challenges like trend, seasonality, and non-constant variance. By the end, you will be equipped to build and evaluate forecasting models tailored to univariate data.

Forecasting with PyTorch Lightning

This module guides learners through building and evaluating multivariate time series forecasting models using PyTorch Lightning. You will explore data preprocessing with TimeSeriesDataSet, implement linear regression, feedforward, and LSTM neural networks, and monitor training progress with TensorBoard. By the end, you'll be able to construct and assess forecasting models for complex time series data.

Global Forecasting Models

This module introduces advanced deep learning techniques for time series forecasting, focusing on multi-step and multi-output predictions using global LSTM models. Learners will discover how to prepare and train models on multiple time series, handle seasonal data, and optimize hyperparameters for improved forecasting accuracy.

Advanced Deep Learning Architectures for Time Series Forecasting

This module delves into advanced deep learning models for time series forecasting, including DeepAR, Transformers, Temporal Fusion Transformers, and Informer architectures. Learners will gain hands-on experience with popular libraries such as GluonTS, NeuralForecast, and PyTorch Forecasting, and learn to optimize model performance through practical techniques like learning rate tuning.

Probabilistic Time Series Forecasting

This module introduces probabilistic approaches to time series forecasting, focusing on methods to quantify and interpret forecast uncertainty. Learners will explore techniques such as exceedance probability, prediction intervals, and advanced neural network models like LSTM and DeepAR, as well as Gaussian Processes. By the end, you'll be able to implement and evaluate probabilistic forecasts to support more informed decision-making.

Deep Learning for Time Series Classification

This module introduces learners to deep learning techniques for classifying time series data, utilizing both PyTorch Lightning and the sktime library. You will gain hands-on experience building data pipelines, implementing convolutional neural networks and ResNets, and exploring alternative frameworks for time series classification tasks.

Deep Learning for Time Series Anomaly Detection

This module introduces deep learning techniques for detecting anomalies in time series data, focusing on models such as autoencoders, LSTMs, and VAEs. Learners will gain hands-on experience building and evaluating these models using Python libraries like PyOD. By the end, you'll understand how to leverage reconstruction and prediction errors to identify unusual patterns in real-world datasets.