Talk at PyData Berlin about Manifold Learning and Applications
4 likes953 views
The document discusses manifold learning and dimensionality reduction techniques presented by Dr. Stefan Kühn at the PyData Conference in Berlin, focusing on the mathematical properties of manifolds and good visualizations. It outlines various manifold learning methods available in sklearn, including locally linear embedding, t-distributed stochastic neighbor embedding, and others. Resources for further exploration, such as GitHub examples and scikit-learn documentation, are also provided.
Talk at PyData Berlin about Manifold Learning and Applications
- 1. Manifold Learning and Dimensionality Reduction for Data Visualization and Feature Engineering Berlin, July 7th, 2018 PyData Conference Berlin Dr. Stefan Kühn https://www.xing.com/profile/Stefan_Kuehn46 https://www.linkedin.com/in/stefan-k%C3%BChn-020a34119/ https://de.slideshare.net/StefanKhn4
- 2. What is a manifold? Mathematical concept from Differential Geometry ManifoldLearningandDataVisualization
- 3. What are properties of a manifold? Important Properties – Topology and more • Number of Connected Components • Holes • Curvature • Smoothness • Dimensionality • …you_name_it… ManifoldLearningandDataVisualization
- 4. What are properties of a good visualization? Preserve important properties • Number of connected components? • Holes? • Curvature? • Smoothness? • Dimensionality? • Distances between points? • Angles, orientations? • Local versus global properties? ManifoldLearningandDataVisualization You cannot have it all!
- 5. Manifold Learning Methods in sklearn • Locally Linear Embedding • Neighborhood-preserving • Isomap • Quasi-isometric • Multi-Dimensional Scaling (MDS) • Quasi-isometric • Spectral Embedding • Spectral clustering based on similarity • T-Distributed Stochastic Neighbor Embedding (tSNE) • Preserves probabilities • Local Tangent Space Alignment (LTSA) ManifoldLearningandDataVisualization
- 6. Local Tangent Space Alignment ManifoldLearningandDataVisualization
- 7. Demo Time Sometimes, words are insufficient… But then God gave us code! ManifoldLearningandDataVisualization
- 8. Resources Scikit-learn documentation http://scikit-learn.org/stable/modules/manifold.html http://scikit-learn.org/stable/auto_examples/manifold/plot_compare_methods.html http://scikit-learn.org/stable/auto_examples/manifold/plot_manifold_sphere.html http://scikit-learn.org/stable/modules/random_projection.html http://scikit-learn.org/stable/modules/generated/sklearn.decomposition.PCA.html Github repo with worked examples https://github.com/cc-skuehn/Manifold_Learning Jupyter Lab https://jupyterlab.readthedocs.io/en/stable/index.html Citation Scikit-learn: Machine Learning in Python, Pedregosa et al., JMLR 12, pp. 2825-2830, 2011. http://jmlr.csail.mit.edu/papers/v12/pedregosa11a.html ManifoldLearningandDataVisualization
- 9. Thank you! Berlin, July 7th, 2018