Book Review: Data Centric Machine Learning with Python

Open original

A review of "Data-Centric Machine Learning with Python" arguing that ML success depends more on data quality than model sophistication, and explaining why this approach has become standard practice. The reviewer extends this to show how LLMs fit into data-centric pipelines as both label generators and sources of weak supervision.

  • The shift to data-centric ML happened years ago through landmark projects like Netflix Prize, ImageNet, and Word2Vec—not a recent invention—driven by the internet's planet-scale data generation.
  • Data preparation is the real bottleneck in modern ML: cleaning, imputing missing values, extracting features, removing bias, and iteratively refining labels through human-in-the-loop active learning takes far more effort than training.
  • LLMs can be treated as replacements for human subject matter experts in labeling and weak supervision, which means the techniques for managing SME bias and inconsistency now apply directly to LLM outputs.

The book argues that ML pipelines get more mileage from high-quality data than from sophisticated models, yet most practitioners obsess over model architecture. The author traces how this shift already happened—Netflix's collaborative filtering work, ImageNet's crowdsourced labeling, Word2Vec—all demonstrated that data quality matters more than people assumed. The internet amplified this trend by letting companies like Netflix and Google generate massive datasets at planetary scale. Early academic ML relied on small hand-labeled datasets from grad students. Modern practice evolved to handle bigger data needs through weak supervision, transfer learning, active learning, and synthetic labels. Now LLMs have complicated things further: they generate labels themselves but carry the same biases and inconsistencies as human experts, making the book's guidance on working with subject matter experts suddenly relevant to everyone managing AI systems. Part 2 of the book covers handling SMEs effectively, which matters because most organizations don't teach ML engineers this skill—but now everyone needs it to manage LLM outputs. Part 3 digs into the practical grind: cleaning messy data, imputing missing values, extracting features, removing bias, and implementing the "data flywheel" where you start with noisy data, build a model, then use active learning with human feedback to iteratively improve both. It covers transfer learning, adversarial learning, and semi-supervised approaches you can actually use. The practical takeaway is that LLMs are powerful enough to solve many problems through prompting alone, but you unlock more value by treating them as components in a traditional ML pipeline—using them to generate labels or weak supervision signals for building scalable models. Part 4 offers organizational guidance for senior roles trying to shift their company's culture toward data-centric thinking, including implementation checklists.

Ask about this article