Loading project…
PythonAdvancedabout 8 hours16 lessons
Recommended first: Orchard Vision
A greenhouse where numbers sprout. You start with a seed: ten hand-designed pen strokes that a generator wobbles into hundreds of handwritten-looking digits. Then you grow a brain for them, one NumPy function at a time: a single neuron, activations, layers as matrix multiplies, softmax and cross-entropy, careful seed weights, backpropagation by hand, a gradient check that catches your own bugs, mini-batch training, a confusion matrix, regularisation, learning-rate seasons and early stopping. A live stage draws the network with weight-coloured edges, the loss and accuracy curves, the digits it gets wrong, and a pad where you draw a digit and your own predict() reads it.
Each lesson adds one working part to your build.
Chapter 1
Draw the digits, then grow a neuron, an activation and a whole layer.
Chapter 2
Score the network, start it sensibly, and learn how blame flows backwards.
Chapter 3
Check your gradients, then train in loops, in batches, and look at the mistakes.
Chapter 4
Overfitting, regularisation, learning-rate seasons, early stopping and the finished reader.