Data Science in Practice

A practical data science course — from the workflow and data preparation to machine learning and putting results into the real world. Aimed at a developer or analyst who wants to learn to extract value from data: to load and clean it, explore it, visualize it, build and evaluate models, and understand where data science ends and engineering begins. Basic Python is enough (the collection has a separate Python course); the whole stack is Python.

The main principle of the course: 80% of a data scientist’s work is the data, not the models. Beautiful algorithms are useless on dirty data; “garbage in, garbage out.” So the emphasis is on preparation, cleaning, exploration, and honest evaluation. The second through-line is skepticism and rigor: correlation isn’t causation, a good metric on the training set means nothing, and a model that “works” in a notebook can fail in production because of data leakage. Data science is the discipline of drawing careful conclusions from noisy reality.

Topics build up: the workflow and tooling, NumPy, pandas, data cleaning, transformation and aggregation, exploratory analysis and visualization, statistics, an introduction to machine learning, scikit-learn, model evaluation and working with features, and putting things into production. The 2026 stack: the core (NumPy, pandas, matplotlib/seaborn, scikit-learn, Jupyter) plus modern tools (Polars, DuckDB) — but the principles matter more than the specific libraries.

Each lesson contains theory, code examples (Python), a checklist of practical tasks, and a quiz. It’s convenient to practice in Jupyter/a notebook on real datasets. This course is the foundation of applied analysis; for the depth of machine learning proper and modern AI, see the ML/AI course, and for LLM applications, see the LLM course.