Essential Readings in Machine Learning | by TDS Editors | Jan, 2023

If one of your goals for the coming year is to expand your knowledge of machine learning, you’re in the right place. Regardless of how experienced you are with algorithms, hyperparameter tuning, or MLOps, you can find multiple entryways into this topic, and our authors—many of whom are seasoned ML experts—have a special knack for translating complex concepts into engaging and actionable posts.

This week, we’ve selected some of the best ML articles we published in recent weeks, including both introductory-level guides and more advanced tutorials and project walkthroughs. Pick one, or read several: you’re bound to learn something new.

• Getting comfortable with ML pipelines. “Implementing machine learning in production environments can be a challenging and time-consuming process for many companies,” says Chayma Zatout—which is why MLOps has become a crucial approach for thinking about machine learning workflows holistically. Read Chayma’s post for an accessible, ground-up introduction to this key topic.
• Decision trees, patiently explained. If you’re eager to get hands-on with an actual model, don’t miss Frauke Albrecht’s debut TDS article. It’s a detailed, step-by-step tutorial that shows you how to build a decision tree for a classification problem, and covers both the underlying math and a Python implementation.

• Looking under the hood of a recommender system. Wen Yang recently won her company’s Innovation Days Award, and her latest post walks us through the project that got her there: a machine learning-based tool that recommends hiking trails for nature lovers. It’s a fascinating exploration of what goes into building and training a model on a tight schedule, as well as a helpful roadmap for doing well at hackathons.
• A (very) thorough guide to feature engineering. Optimizing your data for the model you’re training can be the deciding factor in your project’s success. Dominik Polzer’s excellent deep dive on feature engineering covers seven common approaches (from encoding and vectorizing to feature crossing), and shows how to apply them to real-world datasets.
• Look up to the stars (are they stars, though?). If you love both ML and astronomy, don’t miss Mohammed Saifuddin’s recent article. It tackles the stellar classification problem—how to identify a given celestial body as a star, quasar, or galaxy based on spectral characteristics—with the aid of a thorough (and well-explained) modeling process.
• What’s the best approach for solving business problems with ML? As more and more companies rely on machine learning for crucial tasks like churn prediction, landing on the right approach becomes key. Samuele Mazzanti recently unpacked a particularly common dilemma: whether it’s better to rely on multiple specialized models, or to opt for one general model instead.
• Beyond the machine: exploring rule-based learning. At the core of Khuyen Tran’s tutorial is an important observation: that human intuition and subject-matter expertise can enhance the performance of ML models. She goes on to show how we can leverage the human-learn Python package to achieve a best-of-both-worlds result.

If one of your goals for the coming year is to expand your knowledge of machine learning, you’re in the right place. Regardless of how experienced you are with algorithms, hyperparameter tuning, or MLOps, you can find multiple entryways into this topic, and our authors—many of whom are seasoned ML experts—have a special knack for translating complex concepts into engaging and actionable posts.

This week, we’ve selected some of the best ML articles we published in recent weeks, including both introductory-level guides and more advanced tutorials and project walkthroughs. Pick one, or read several: you’re bound to learn something new.

• Getting comfortable with ML pipelines. “Implementing machine learning in production environments can be a challenging and time-consuming process for many companies,” says Chayma Zatout—which is why MLOps has become a crucial approach for thinking about machine learning workflows holistically. Read Chayma’s post for an accessible, ground-up introduction to this key topic.
• Decision trees, patiently explained. If you’re eager to get hands-on with an actual model, don’t miss Frauke Albrecht’s debut TDS article. It’s a detailed, step-by-step tutorial that shows you how to build a decision tree for a classification problem, and covers both the underlying math and a Python implementation.

• Looking under the hood of a recommender system. Wen Yang recently won her company’s Innovation Days Award, and her latest post walks us through the project that got her there: a machine learning-based tool that recommends hiking trails for nature lovers. It’s a fascinating exploration of what goes into building and training a model on a tight schedule, as well as a helpful roadmap for doing well at hackathons.
• A (very) thorough guide to feature engineering. Optimizing your data for the model you’re training can be the deciding factor in your project’s success. Dominik Polzer’s excellent deep dive on feature engineering covers seven common approaches (from encoding and vectorizing to feature crossing), and shows how to apply them to real-world datasets.
• Look up to the stars (are they stars, though?). If you love both ML and astronomy, don’t miss Mohammed Saifuddin’s recent article. It tackles the stellar classification problem—how to identify a given celestial body as a star, quasar, or galaxy based on spectral characteristics—with the aid of a thorough (and well-explained) modeling process.
• What’s the best approach for solving business problems with ML? As more and more companies rely on machine learning for crucial tasks like churn prediction, landing on the right approach becomes key. Samuele Mazzanti recently unpacked a particularly common dilemma: whether it’s better to rely on multiple specialized models, or to opt for one general model instead.
• Beyond the machine: exploring rule-based learning. At the core of Khuyen Tran’s tutorial is an important observation: that human intuition and subject-matter expertise can enhance the performance of ML models. She goes on to show how we can leverage the human-learn Python package to achieve a best-of-both-worlds result.