Learn how to build a machine learning pipeline entirely from scratch, using software engineering best practices.

Building a Machine Learning Pipeline from Scratch.png

docker_setup.tar.gz

omegaconf-code-widget

omegaconf-spa-widget

omegaconf-spa-widget-copy

omegaconf-code-widget-copy

omegaconf_example

dataset_model_examples

logging_example

docstring_example

pipeline_iris_classification

pipeline_autompg_regression

reporting

pipeline_packaging

inference

unit_testing

pipeline_unit_testing

experiment_tracking

argparse_example

Machine learning (ML) has matured into a mainstream development activity, and data scientists are expected to be able to write production-grade training pipelines. This course will provide you with a foundation in ML pipeline development guided by best practices in software engineering.

You’ll start by learning about code organization, style, and conceptual ideas, such as topological sorting of directed acyclic graphs. You’ll dive into the hands-on development of an ML pipeline. You’ll learn some advanced Python concepts and useful libraries. Finally, you’ll learn about testing methodologies and how to package your code into a distributable library.

By the end of this course, you’ll understand what makes a great ML pipeline work, better appreciate software engineering processes, improve your Python knowledge, and learn how to build ML training pipelines. This knowledge will make you more attractive to potential employers and improve your chances of thriving as a new data scientist.

Building a Machine Learning Pipeline from Scratch

**(True or False)** Observability is critical for model training and deployment.

(True or False) Observability is critical for model training and deployment.

**(True or False)** Observability is critical for monitoring data quality and model performance after the model has been deployed.

(True or False) Observability is critical for monitoring data quality and model performance after the model has been deployed.

What impact does a data distribution shift have on the ML life cycle?

What impact does a data distribution shift have on the ML life cycle?

**(True or False)** A model trained on a dataset a second time will always have the exact performance as the first model.

(True or False) A model trained on a dataset a second time will always have the exact performance as the first model.

**(Select all that apply.)** Which options help us achieve reproducibility?

(Select all that apply.) Which options help us achieve reproducibility?

Test yourself on the concepts learned in this chapter.

Introduction

Getting Started

Structuring the ML Pipeline

Directed Acyclic Graphs (DAGs)

The ML Library

The Pipeline Core

Extending the Pipeline

Testing

Deployment

Other Considerations

Wrapping Up

Appendix

Final Assessment

Quiz: Observability and Reproducibility