Foundations
What machine learning is, how to look at and prepare data, and how to judge a model honestly.
About 4 hours of study
Learning objectives
- Use the vocabulary of machine learning
- Summarize and scale features
- Detect overfitting with held-out data
Lessons
- 1
What machine learning is
The idea behind machine learning: instead of writing the rules, let the computer find them from examples. Training and prediction, features and labels, and the three kinds of learning.
- 2
Data, statistics and features
Look at the numbers before any model: mean, median and standard deviation, outliers, scaling features with min-max and z-scores, one-hot encoding and missing values.
- 3
Training, testing and overfitting
Why a model must be judged on data it never saw: train and test sets, overfitting and underfitting, bias and variance, validation sets and k-fold cross-validation.
Module practice
- Score a cut-off ruleEasy
- Learn the best cut-offMedium
- The OneR classifierHard
- Mean, median and spreadEasy
- Min-max scalingMedium
- Prepare a feature tableHard
- Hold out every k-th exampleEasy
- k-fold cross-validation foldsMedium
- Memorizer against a simple ruleHard