Watching lessons is only half of learning. These four projects turn the maths from the course into working code: you build a well-known machine learning method from scratch in Python, using only NumPy or Pandas, so you can see exactly where the vectors, gradients, probabilities and loss functions come in.
Each project comes with a video that walks through the build. Watch it, code along, then try the same thing again on a dataset of your own. Each project also lists the modules it uses, so you can go back and revise anything that feels shaky.
Code linear regression in Python with NumPy and fit it by gradient descent, with no machine learning library. Then try it on your own data and watch the loss fall as you change the learning rate. Uses modules 4 to 6.
Video 14minImplement principal component analysis in Python with NumPy: centre the data, build the covariance matrix, find its eigenvectors and project the data down to fewer dimensions. Uses modules 2, 3 and 9.
Video 17minStart a series that codes a Naive Bayes classifier from scratch in Python on the Titanic dataset, beginning with the data and a recap of how the algorithm uses Bayes' theorem. Uses modules 7 and 11.
Video 11minCode a neural network in Python with NumPy and Pandas, including the forward pass, backpropagation and training. This one pulls together matrices, the chain rule and optimization. Uses modules 2, 5 and 6.
Video 35min