Introduction

---

• The TensorFlow machine learning library from Google provides many algorithms and models for efficient pipelines to process and forecast large time-series datasets.

Feature Engineering

---

• Use sine and cosine transformations to represent the periodic or cyclical nature of time-series data.

Data Windowing and Making Datasets

---

• Data windows enable single and multi-step time-series forecasting.

Single Step Forecasts

---

• Use the keras API to define neural network layers and attributes to construct different machine learning pipelines.

Multi Step Forecasts

---

• If a label_columns argument is not provided, the data window will forecast all features.