Clustering, neural networks and practice
Finding groups without labels, from perceptrons to neural networks, and running a real project responsibly.
About 5 hours of study
Learning objectives
- Cluster data with k-means
- Train a perceptron and describe neural networks
- Run a leak-free, fair workflow
Lessons
- 1
Clustering with k-means Intermediate
Find groups in data without labels: the k-means loop of assigning points and moving centroids, inertia, local optima and k-means++, and choosing k with the elbow method.
- 2
Perceptrons and neural networks Intermediate
From one artificial neuron to a network: the perceptron and its learning rule, why one neuron cannot learn XOR, hidden layers, activation functions and backpropagation.
- 3
Machine learning in practice Intermediate
The work around the algorithm: the project workflow, baselines, the scikit-learn libraries, data leakage, monitoring for drift, and responsible machine learning with checks for each group of people.
Module practice
- Assign points to the nearest centroidEasy
- k-means on a lineMedium
- k-means in two dimensions with inertiaHard
- A neuron with a step activationEasy
- Train a perceptronMedium
- Forward pass of a small neural networkHard
- A majority-class baselineEasy
- Accuracy and recall for each groupMedium
- A leak-free prediction pipelineHard