This talk will present a recent approach of geometrical understanding of data, called Topological Data Analysis. It will focus on three algorithms based on manifold learning and algebraic topology: Persistent Homology, UMAP and Mapper. We will show how do they work and use its Python implementations to exhibit cases of use for clustering, classification and recommendation in real life problems.
The objective of this talk is to present a recent approach of geometrical understanding of data, called Topological Data Analysis (TDA).
It will be specially focused on three algorithms based in manifold learning and algebraic topology: Persistent Homology, UMAP and Mapper. These techniques provide low dimensional visualizations of data aiming to preserve its topological structure. We will show how these algorithms work and compare them with other classical techniques of dimensional reduction and data visualization such as PCA, t-SNE, etc.
These algorithms have been implemented as (scikit-learn compatible) Python libraries. We will exhibit in Jupyter Notebooks some code examples of usage for clustering, classification and recommendation in real life problems. In addition, we will present some examples using data from Properati. Since its datasets are open to the community, the audience will be able to replicate all the examples.
This talk is self-contained, no previous knowledge of topology is needed.