Data Science Roadmap

Variables, lists, dictionaries, functions, loops — in a notebook or the console where feedback is instant. Python is the field's shared language; fluency here pays interest at every later stage.

Distributions, mean versus median, variance, sampling, and above all correlation versus causation. Most real-world data mistakes are statistical reasoning mistakes, not coding mistakes.

Before touching any dataset, practice writing down the question: what would change if I knew this answer? Data work without a sharp question produces charts, not insight.

Files, databases, APIs, exports — and every one of them lies a little. Knowing how data was collected tells you what its blind spots are before a single row is loaded.

Loading, filtering, grouping, joining, reshaping. Wrangling is 60–80% of real data work, so being fast here changes what projects are feasible for you.

Missing values, duplicates, inconsistent formats, outliers. Cleaning is not a chore before analysis — the choices you make here are analytical decisions that change your conclusions.

Most of the world's data sits in SQL databases, and most data jobs test for it. SELECT, JOIN, GROUP BY and window functions cover the large majority of daily use.

Learn one plotting library well and the grammar of good charts: what comparison the chart makes, and whether it is honest. A well-chosen chart is an argument; make yours truthful.

Pick a dataset you care about, pose a question, wrangle, explore, visualize, and write a short conclusion — including what you cannot conclude. This single project exercises everything in this stage.

Train/test splits, overfitting, and the intuition behind a few core algorithms (linear models, trees, ensembles). The concepts transfer to every model you will ever use; the specific algorithm matters less than knowing why it fits.

Accuracy is often the wrong metric; leakage is everywhere; baselines are humbling. A simple model honestly evaluated beats an impressive model with inflated numbers — in interviews and in production.

Analytics (insight, dashboards, decisions) and machine learning engineering (models in production) are different jobs with a shared foundation. Try both flavors in small projects before you decide where to go deep.

The analysis that changes a decision is the one someone understood. Practice the two-minute version: question, answer, confidence, caveat — most stakeholders will never read your notebook.

Real data, a real question, honest conclusions, readable write-up. Employers and collaborators read your reasoning, not your code first.

Kaggle-style competitions teach evaluation discipline and expose you to other people's solutions — worth it even if you place mid-table. Do not let leaderboard chasing replace real projects.

Data skills multiply when combined with a field — finance, health, games, climate, anything. The analyst who understands the domain asks questions the pure technician cannot.

For each project: the question, your approach, what surprised you, what you would do differently. That narrative is your actual portfolio; the notebooks are appendices.