Understanding machine learning through beautiful algorithm animations

Here we present beautiful **animated visualizations** for some popular **Machine Learning** algorithms, built with the `R`

package `animation`

. These animations help to understand algorithm iterations and hyper-parameter tuning.

The source code is available on **GitHub**.

Classification **Decision Boundary** of the **Gradient Boosting Machine** (GBM) from the `R`

package `xgboost`

as boosting iterations proceed.

Analysing how the the number of nearest neighbors \(k\) affects the classification **Decision Boundary** of the **KNN** algorithm from the `R`

package fastknn. We consider 2 probability estimators for the class membership probabilities: a *voting rule* and a *weighted voting rule* (shrinkage estimator).

**Gaussian Mixture Model** (GMM) fitted by **Expectation-Maximization** (EM) algorithm with random initialization.

* Source Code* is not available yet.

**Density** estimation using a GMM with 7 components. Model fitting is performed by **Expectation-Maximization** algorithm with randomly assigned initial parameters. Positive definiteness of covariance matrix is achieved replacing the unrestricted maximum likelihood estimator by *Ledoit-Wolf* **shrinkage estimator**.

* Source Code* is not available yet.

Classification **Decision Boundary** of a *Single Layer Feedforward Network* (with 150 randomly assigned hidden neurons) trained by **Extreme Learning Machine** (ELM) algorithm considering ridge regression instead of ordinary least squares estimation. The larger the penalty parameter, the greater the amount of smoothing.

Image pixels grouping into k different clusters using the k-means algorithm. A different color is assigned for each cluster. This simulation is based on the following post: R-bloggers.

Image reconstruction using the k first principal components (PCs).

Developed by David Pinto