01

Featured algorithm

unsupervised · clustering
02

Taxonomy constellation

Every model, mapped by family. Hover a node to light up its branch.

hover a node · it floats
03

The catalogue

Supervised

Supervised learning

Linear Regression

Supervised · Regression

Regression

hover ⤿ math
Supervised learning

ŷ = w · x + b

Learns a best‑fit line to predict continuous values.

MSE lossclosed-forminterpretable

Linear Regression

Supervised learning

Logistic Regression

Supervised · Classification

Classification

hover ⤿ math
Supervised learning

p = σ(w · x + b)

Models class probability using a sigmoid/softmax decision boundary.

log-losssigmoidprobabilistic

Logistic Regression

Supervised learning

Support Vector Machine (SVM)

Supervised · Classification

Classification

hover ⤿ math
Supervised learning

min ½‖w‖² s.t. yᵢ(w·xᵢ+b) ≥ 1

Finds a maximum‑margin separating hyperplane (kernelizable).

max-marginkernelssupport vectors

Support Vector Machine (SVM)

Supervised learning

K‑Nearest Neighbors (KNN)

Supervised · Classification/Regression

Classification/Regression

hover ⤿ math
Supervised learning

ŷ = vote{ y : x ∈ Nₖ(x) }

Predicts from the labels/values of the nearest points in feature space.

lazydistancenon-parametric

K‑Nearest Neighbors (KNN)

Supervised learning

Decision Tree

Supervised · Tree‑based

Tree‑based

hover ⤿ math
Supervised learning

split → max information gain

Learns hierarchical if‑then splits to predict class or value.

Gini / entropyif-theninterpretable

Decision Tree

Supervised learning

Random Forest

Supervised · Ensemble

Ensembles

hover ⤿ math
Supervised learning

ŷ = mode(T₁ … T_B)

Bagging ensemble of trees; reduces variance and improves robustness.

bagginglow variancerobust

Random Forest

Supervised learning

Gradient Boosting

Supervised · Ensemble

Ensembles

hover ⤿ math
Supervised learning

Fₘ = Fₘ₋₁ + ν · hₘ

Builds learners sequentially to correct errors (e.g. XGBoost/LightGBM style).

sequentialresidual fitXGBoost

Gradient Boosting

Supervised learning

MLP (Neural Network)

Supervised · Neural Network

Neural Networks

hover ⤿ math
Supervised learning

a⁽ˡ⁾ = σ(W a⁽ˡ⁻¹⁾ + b)

Learns non‑linear mappings using layered neurons and backpropagation.

backpropReLUuniversal

MLP (Neural Network)

Unsupervised

Unsupervised learning

K‑Means

Unsupervised · Clustering

Clustering

hover ⤿ math
Unsupervised learning

argmin Σ ‖x − μₖ‖²

Partitions data into k clusters by iteratively updating centroids.

Lloyd's algok chosenfast

K‑Means

Unsupervised learning

DBSCAN

Unsupervised · Density‑based Clustering

Clustering

hover ⤿ math
Unsupervised learning

core: |Nε(p)| ≥ minPts

Finds dense regions and marks sparse points as noise/outliers.

densityfinds noiseno k

DBSCAN

Unsupervised learning

PCA

Unsupervised · Dimensionality Reduction

Dimensionality Reduction

hover ⤿ math
Unsupervised learning

max Var(Xw), ‖w‖ = 1

Projects data onto principal components maximizing variance.

eigenvectorsvariancelinear

PCA

Unsupervised learning

t‑SNE

Unsupervised · Dimensionality Reduction

Dimensionality Reduction

hover ⤿ math
Unsupervised learning

min KL(P ‖ Q)

Visualizes high‑dimensional data by preserving local neighborhoods.

local structurenon-linearviz

t‑SNE

Reinforcement

Reinforcement learning

Q‑Learning

Reinforcement · Value‑Based

Value‑Based

hover ⤿ math
Reinforcement learning

Q ← Q + α[ r + γ·maxQ′ − Q ]

Learns state‑action values to choose actions that maximize reward.

off-policyvalue-basedtabular

Q‑Learning

Reinforcement learning

DQN

Reinforcement · Deep RL

Deep RL

hover ⤿ math
Reinforcement learning

L = ( r + γ·maxQ(s′;θ⁻) − Q(s,a;θ) )²

Approximates Q‑values with a neural network for high‑dim inputs.

deep RLreplaytarget net

DQN

Reinforcement learning

PPO

Reinforcement · Actor‑Critic

Actor‑Critic

hover ⤿ math
Reinforcement learning

max E[ min(rₜAₜ, clip(rₜ)·Aₜ) ]

Stabilizes policy updates using a clipped objective (widely used in practice).

on-policyclippedstable

PPO

04

Interactive playground

You are the query point. K-Nearest Neighbours finds your 5 closest points and votes on the class — live.

move your cursor — you are the query
predicted
k=5 vote