algorithms.live

    Machine learning, visualized

    Change the data, the learning rate or the regularisation and watch the model respond. The maths is shown next to the picture, not instead of it.

    Start with the ML landscape and the linear algebra, probability and calculus refreshers, then regression and evaluation before the unsupervised methods.

    Start with The Machine Learning Landscape 29 topics

    Start here

    Beginner topics, in the order they build on each other.

    1. The Machine Learning LandscapeThree independent axes place almost any ML system: supervision, batch versus online, instance versus model-based. Plus the two things that actually sink projects — a poll of ten million people that called an election wrong, and a validation setup that cannot tell overfitting from a data mismatch.Beginner18m
    2. Linear Algebra 3DInteractive 3D visualization of vectors, matrix transformations, and eigenvalues - the mathematical foundation of all machine learning.Beginner20m
    3. Probability Distributions 3DExplore 2D Gaussian distributions, correlation, and mixture models in beautiful 3D visualizations - the foundation of probabilistic ML.Beginner18m
    4. Calculus 3D (Derivatives & Gradients)Master calculus concepts in 3D - derivatives, partial derivatives, gradients, and the chain rule. The mathematical foundation of backpropagation.Beginner22m
    5. K-Nearest Neighbors (KNN) 3DThe simplest ML algorithm - classify points based on their nearest neighbors. Watch majority voting in action in 3D space!Beginner18m
    6. Naive Bayes ClassifierProbabilistic classifier based on Bayes' Theorem! See how prior probabilities and likelihoods combine to make predictions. The foundation of spam detection and text classification.Beginner20m
    7. Linear RegressionThe simplest ML model - learn to fit a line through data points and understand the foundation of all machine learning.Beginner15m
    8. Logistic RegressionFoundation of binary classification - learn decision boundaries, sigmoid function, and probability predictions.Beginner18m
    9. Confusion Matrix & MetricsMaster model evaluation! Learn precision, recall, F1-score, and when accuracy lies. Essential for understanding classification performance.Beginner20m
    10. Cross-ValidationEssential model evaluation technique! Learn K-Fold, Stratified K-Fold, and Time Series Split. Understand how to properly assess model performance and avoid data leakage.Beginner18m
    11. Feature ScalingCritical preprocessing step! Learn why scaling is essential for SVM, KNN, and neural networks. Compare StandardScaler, MinMaxScaler, and RobustScaler.Beginner15m
    12. Decision TreeThe foundation of Random Forests and XGBoost! Learn how decision trees recursively split data using information gain, Gini impurity, and entropy. Watch the tree grow and classify data interactively.Beginner20m
    13. Gradient DescentInteractive 3D visualization of how gradient descent optimizes functions by iteratively moving towards the minimum.Beginner15m
    14. Loss FunctionsCompare different loss functions (MSE, Cross-Entropy, etc.) and understand their use cases.Beginner15m
    15. ConvNetJS 2D ClassificationTrain neural networks directly in your browser with ConvNetJS. Visualize how the decision boundary evolves as the network learns to separate 2D classification problems.Beginner20m

    Next

    Intermediate topics.

    1. Bias-Variance TradeoffTHE most critical concept in ML - understand why models fail through underfitting vs overfitting.Intermediate20m
    2. Learning CurvesCross-validation tells you a model generalises badly. Learning curves tell you which kind of badly — and therefore whether more data will help or waste your money. Every curve here comes from actually refitting a polynomial at each training set size.Intermediate20m
    3. ROC Curve & AUCVisualize ROC curves and AUC, and learn where they beat accuracy — and where they quietly flatter a model. Under a rare positive class the curve still looks excellent while precision collapses, which is why Geron's rule is to prefer the precision-recall curve whenever the positive class is rare.Intermediate22m
    4. Precision/Recall Trade-off & PR vs ROCA classifier outputs a score, not a class — something compares it to a threshold. Drag that threshold and watch every metric move, then drag the positive class rare and watch the ROC curve hold steady while the PR curve collapses. Same model, same scores, two different verdicts.Intermediate22m
    5. Ensemble Methods: Random Forest & BoostingTHE most powerful ML algorithms in production! See how combining weak learners creates strong models. Compare Bagging (Random Forest) vs Boosting (AdaBoost, XGBoost). Watch ensemble magic!Intermediate25m
    6. Overfitting & RegularizationInteractive demonstration of overfitting and how regularization techniques (L1, L2, Dropout) help.Intermediate20m
    7. Loss Landscape 3DTHE most important optimization visualization! Explore loss surfaces in 3D weight space - see convex bowls, non-convex landscapes with multiple minima, saddle points, and ravines. Watch gradient descent navigate different terrains.Intermediate25m
    8. The Curse of DimensionalityIn a unit square a random point is almost never near a border; in a 10,000-dimensional one it almost always is. Two points inside a unit hypercube can sit 408 units apart. Every figure here is computed live from the dimension count, not quoted.Intermediate18m
    9. K-Means Clustering 3DWatch K-Means clustering algorithm discover natural groupings in 3D data - the most popular clustering algorithm in action!Intermediate18m
    10. DBSCAN — Density-Based Clusteringk-means asks how many clusters there are and finds exactly that many, all roughly spherical. DBSCAN asks neither — it finds continuous regions of high density, discovers the count itself, and labels what is left over as anomalies. The real algorithm runs live on two moons.Intermediate20m
    11. PCA (Principal Component Analysis) 3DVisualize dimensionality reduction in 3D - see how PCA finds directions of maximum variance and projects high-dimensional data to lower dimensions.Intermediate20m
    12. Support Vector Machine (SVM) 3DVisualize how SVMs find the optimal decision boundary with maximum margin separation between classes in 3D space.Intermediate20m

    Advanced

    For when the rest feels solid.

    Learn it as part of a path

    AI Engineer: Zero to HeroMath & ML foundations → deep learning → the full LLM application pipeline → production AI systems.ML Engineer: Zero to HeroMath → classical ML in depth → deep learning → designing, deploying and operating ML systems in production.

    Related topics

    Deep LearningML Systems DesignAI Engineering