Pick a folder. Pull out a lesson.

Linear RegressionMinimizing orthogonal Euclidean residuals in parameter space. Closed-form normal equations versus iterative gradient descent steps.

Logistic RegressionProjecting continuous feature combinations onto calibrated probability bounds. Analyzing logit slopes, odds ratios, and linear decision boundaries.

Confusion Matrix & F1Navigating severe class imbalance where baseline accuracy fails. Harmonic balance between false positive alarms, missed detections, and F1 optimization.

ROC & AUCMapping the sensitivity vs. specificity tradeoff across continuous decision thresholds with confusion matrix projections and integral AUC estimation.

Cross-ValidationMitigating sample bias and estimating performance variance through rotational holdout splits and out-of-fold validation.

Trees & Random ForestsRecursive feature space partitioning, Gini impurity minimization, and bootstrap aggregation (bagging) with random feature sub-sampling to suppress variance.

Neural NetworksConstructing feed-forward computational graphs from first principles: layer activations, synaptic weights, Adam optimization, and live backpropagation feedback loops.

Convolutional NetworksSpatial receptive fields, 2D discrete convolution kernels, hierarchical feature maps, and spatial invariance through pooling layers.

Recurrent Networks & LSTMsSequential temporal dynamics, Backpropagation Through Time (BPTT), vanishing gradients, LSTM memory gates, and Transformer self-attention.

Random ForestsBagging, random feature sub-sampling, and majority voting. How averaging many decorrelated trees collapses variance without adding bias.

Bias-VarianceDecomposing expected error into bias squared, variance, and irreducible noise. Underfitting, overfitting, and the KNN knob that dials between them.

Train / Test / ValidationWhy one dataset must become three, what each partition is actually for, and how a single careless peek turns your final number into fiction.

Double DescentError that rises at the interpolation threshold and then falls again. Why over-parameterised models generalise, and what minimum-norm bias has to do with it.

Equality of OddsDefining and measuring parity in true and false positive rates across groups, the equal-opportunity relaxation, and the three stages at which you can intervene.

Reinforcement LearningLearning from delayed consequences instead of labels. Markov decision processes, the explore-exploit dilemma, grid worlds, and the Bellman recursion.

The TransformerEliminating recurrence with constant-path multi-head self-attention. Interactive dual-tower architecture, coreference attention simulator, and autoregressive generation.