Andrei Erhan
Lead Data Scientist · AI/ML Team Lead
I lead a data science team at maib and teach people how machine learning actually works. This site is my map of the field: every topic comes with a playground you can train in your browser.
Every model in one idea
Thirteen models, one cloud of 360 stars. Each model reshapes the same points, so you can see what it actually does. The clustering scenes run the real algorithms.
Linear regression
Draws the straight line that best predicts a number.
The map of machine learning
Eight topics, eight planets. Start at Data preparation and follow the orbits outward. The full map on the Learn page adds two overview stops, supervised and unsupervised learning.
Train it yourself
Pick a dataset, press Train, scrub through the epochs, and compare up to four models side by side. Everything runs in your browser.
Regression
SupervisedFit curves with 8 models and watch over- and under-fitting happen.
Classification
SupervisedDecision boundaries for logistic regression, trees, forests, SVMs and an MLP.
Clustering
Unsupervisedk-means, e-means, DBSCAN, Gaussian mixtures and agglomerative clustering.
Dimensionality reduction
UnsupervisedUnroll a swiss roll with PCA, Kernel PCA, t-SNE and UMAP.
Embeddings lab
RepresentationType sentences and see a real transformer place them in space.
RAG lab
SystemsChunk, embed, retrieve, rerank and generate, one step at a time.
Recommender lab
SystemsRate a few titles and watch matrix factorization find your taste.