AI & Data
Machine Learning Essentials
Build, evaluate and deploy models you can defend. Twelve weeks, heavy on the evaluation.
Overview
A working introduction to machine learning that spends as much time on whether a model is any good as on building it. You will train models, then take them apart: where the accuracy came from, what the model does on data it has never seen, and when the honest answer is that there is not enough signal.
Places are limited and applications are reviewed, because the cohort works better when everyone can keep pace.
Learning outcomes
Frame a business problem as a supervised learning task, or recognise when you cannot
Train regression and classification models with scikit-learn
Evaluate a model properly: train/test splits, cross-validation, the right metric for the problem
Recognise and fix overfitting, leakage and imbalanced classes
Engineer features from raw data
Serve a trained model behind a simple API
Curriculum
01Week 1-2: Foundations
- What machine learning can and cannot do
- Supervised, unsupervised and the difference that matters
- The train/test split and why it is not optional
02Week 3-4: Regression
- Linear and polynomial regression
- Regularisation
- Residuals and what they tell you
03Week 5-6: Classification
- Logistic regression, decision trees, random forests
- Precision, recall and the cost of each error
- Imbalanced classes
04Week 7: Evaluation in depth
- Cross-validation
- Data leakage, and how it fakes a good score
- Baselines you must beat
05Week 8-9: Features and tuning
- Feature engineering and selection
- Scaling and encoding
- Hyperparameter search
06Week 10: Putting a model to work
- Saving and loading models
- A minimal prediction API
- Monitoring and drift
07Week 11-12: Capstone
- End-to-end project on a dataset of your choice
- Written evaluation of your own model's weaknesses
- Presentation and review
Prerequisites
- Confident Python, including pandas
- School-level statistics: mean, variance, probability
- Completion of Data Analytics with Python and Power BI, or equivalent experience