Getting Started (Python Notebooks)

Note

For a complete list of tutorials using Python notebooks, see the Amazon Forecast Github Samples page.

To get started using Amazon Forecast APIs with Python notebooks, see the Getting Started Tutorial. The tutorial guides you through the core steps of Forecast from start to finish.

For basic tutorials for specific processes, refer to the following Python notebooks:

1. Preparing data - Prepare a dataset, create a dataset group, define the schema, and import the dataset group.

2. Building your predictor - Train a predictor on the data you imported into your Forecast dataset.

3. Evaluating predictors - Obtain predictions, visualize predictions, and compare results.

4. Retraining predictors - Retrain an existing predictor with updated data.

5. Upgrade to AutoPredictor - Upgrade legacy predictors to AutoPredictor.

6. Clean Up - Delete the dataset groups, predictors, forecasts created during the tutorials.

To repeat the Getting Started tutorial with AutoML, see Getting Started with AutoML.

Advanced Tutorials

For more advanced tutorials, refer to the following Python notebooks:

• Item-level Explainability - Understand how dataset attributes impact forecasts for specific time series and time points.

• Comparing multiple models - Create predictors using Prophet, ETS, and DeepAR+, and compare their performances by visualizing the results.

• Cold start forecasting - Use item metadata and the DeepAR+ algorithm to forecast for cold-start scenarios (when there is little to no historical data).

• Incorporating related time-series datasets - Use related time-series datasets to improve the accuracy of your model.

• Incorporating item metadata - Use item metadata to improve the accuracy of your model.

• Using the Weather Index - Use the Weather Index to incorporate historical and projected weather information when training your predictors.

• Performing What-if analysis - Explore different pricing scenarios and evaluate how it impacts demand.

• Evaluate item-level accuracy - Export backtest metrics and forecasts, and evaluate the item-level performance of your predictor.