Data Science Practice Hub

Most data science content explains algorithms; far less explains the working practice — the open-source tooling, pipeline hygiene, and statistical discipline that decide whether an analysis survives contact with production. This cluster covers that practice layer: the dataframe engines, validation contracts, experiment trackers, and testing habits used by teams (human and agent alike) shipping data science inside real projects. For the underlying theory — model families, evaluation metrics, deep learning — see the ML Hub.

Working with data: engines and query layers

The daily tools for wrangling tabular data, and how to choose between them.

• Pandas vs Polars: Choosing a DataFrame Engine — eager vs lazy execution, the expression API, and migration paths
• DuckDB for In-Process Analytics — SQL over Parquet and dataframes without a server, and larger-than-memory queries

Modeling workflow: pipelines, boosting, tracking

From features to fitted model, with the guardrails that keep results honest.

• Scikit-Learn Pipelines and Leakage-Safe Preprocessing — ColumnTransformer, fit/transform boundaries, and cross-validating the whole pipeline
• Gradient Boosting in Practice: XGBoost, LightGBM, CatBoost — early stopping, categorical handling, and a tuning order that converges
• Experiment Tracking with MLflow — runs, artifacts, the model registry, and self-hosted deployment

Trustworthy results: validation, reproducibility, experiments

The discipline that separates evidence from noise.

• Data Validation with Pandera and Great Expectations — executable schema contracts at pipeline boundaries
• Reproducible Data Science: Environments, Seeds, and Data Versioning — uv lockfiles, random-seed control, and DVC
• A/B Testing in Practice — power analysis, peeking, CUPED, and trustworthiness checks

Adjacent Hubs

• ML Hub — model families, deep learning, and evaluation theory
• Data Engineering Hub — the pipelines and warehouses feeding these workflows
• ML Model Deployment Hub — serving, monitoring, and MLOps at deployment time
• Operations Research Hub — optimization and simulation practice