A-CUBE

// Intermediate — some coding experience

Data Science

Build models that work outside a notebook

From Python and statistics through to machine-learning models you deploy and monitor. For people who can already code a little and want the real thing.

Python pandas NumPy scikit-learn SQL FastAPI Docker Git MLflow

What you will build, week by week

7 modules, every one of them hands-on.

Download as PDF

  1. 01 Module 1 — Python and data handling at speed

    3 weeks

    Getting your Python and your data wrangling to a professional standard.

    • Python beyond the basics: comprehensions, generators, typing
    • NumPy arrays and vectorised thinking
    • pandas for serious datasets: reshaping, merging, grouping
    • SQL for pulling your own training data
    • Git and code review from day one

    Hands-on Build a reusable data-cleaning pipeline for a messy public dataset.

  2. 02 Module 2 — Statistics and probability you will actually use

    3 weeks

    The foundation that separates a data scientist from someone calling library functions.

    • Probability, distributions and expectation
    • Hypothesis testing and p-values, and their honest limits
    • Confidence intervals and uncertainty
    • Bayesian thinking, informally
    • Experiment design and A/B testing

    Hands-on Design and analyse an experiment, reporting uncertainty properly.

  3. 03 Module 3 — Exploratory analysis and feature engineering

    3 weeks

    Most model performance is won here, not in the algorithm.

    • Systematic exploratory data analysis
    • Handling missing data, outliers and skew
    • Encoding categorical variables sensibly
    • Scaling, binning and creating features that carry signal
    • Leakage: how it happens and how to catch it

    Hands-on Take a raw dataset to a modelling-ready feature set, documented.

  4. 04 Module 4 — Machine learning, supervised

    5 weeks

    The core algorithms, how they fail, and how to tell whether yours is any good.

    • Linear and logistic regression, properly understood
    • Decision trees, random forests, gradient boosting
    • Train/validation/test discipline and cross-validation
    • Metrics: accuracy, precision, recall, ROC-AUC, and choosing between them
    • Overfitting, regularisation and the bias–variance tradeoff
    • Class imbalance in real business data

    Hands-on Build a classifier for a real problem and defend your metric choice.

  5. 05 Module 5 — Unsupervised learning and time series

    3 weeks

    When there is no label to learn from, and when time changes everything.

    • Clustering: k-means, hierarchical, DBSCAN
    • Dimensionality reduction with PCA
    • Customer segmentation as a business problem
    • Time-series forecasting and its particular traps

    Hands-on Segment a customer base and present what each segment is worth.

  6. 06 Module 6 — Putting a model into production

    4 weeks

    The part most courses skip, and the part employers most want.

    • Serving a model behind a FastAPI endpoint
    • Containerising with Docker
    • Model versioning and experiment tracking with MLflow
    • Monitoring: drift, degradation and when to retrain
    • Cost, latency and the engineering constraints that shape modelling

    Hands-on Deploy your model as a running API your classmates can call.

  7. 07 Module 7 — Capstone and interviews

    3 weeks

    One substantial project, taken all the way, plus interview preparation.

    • Scoping a capstone from a real dataset
    • Building, evaluating, deploying and documenting it
    • Writing up results for a technical and a business audience
    • Mock interviews: statistics, ML theory, SQL, and a case study

    Hands-on A deployed, documented capstone project in your own GitHub repository.

Before you join

Comfortable writing basic Python — loops, functions, lists and dictionaries — and school-level mathematics. If you are starting from zero, take Data Analysis first; it leads into this course.