Parameters of t SNE
Download as PPTX, PDF0 likes163 views
The document discusses the parameters of t-SNE (t-distributed stochastic neighbor embedding) for dimensionality reduction, specifically focusing on perplexity and the number of iterations required for convergence. It suggests a perplexity range of 5 to 50, using the UCI ML digit image dataset, and notes the need for various iterations for the algorithm's convergence. Key suggested values include perplexity options of 5, 30, and 100, with n_iter settings of 250, 500, and 2000.
Parameters of t SNE
- 2. t-SNE preserves neighbourhood 2 Dimension 1 Dimension2 Mapping that preserves local structure of data P-dimensional space 2D space Neighbors remain neighbors t-SNE: t-distributed Stochastic Neighbor Embedding
- 3. 3 Perplexity (perplexity): somehow shows the number of close neighbors each point has. ● suggested range: between 5 and 50. Dataset: UCI ML digit image data Parameters of t-SNE (Perplexity)
- 4. 4 Perplexity (perplexity): somehow shows the number of close neighbors each point has. ● suggested range: between 5 and 50. Perplexity=5 Dataset: UCI ML digit image data Parameters of t-SNE (Perplexity)
- 5. 5 Perplexity (perplexity): somehow shows the number of close neighbors each point has. ● suggested range: between 5 and 50. Perplexity=5 Perplexity=30 Dataset: UCI ML digit image data Parameters of t-SNE (Perplexity)
- 6. Parameters of t-SNE (Perplexity) 6 Perplexity (perplexity): somehow shows the number of close neighbors each point has. ● suggested range: between 5 and 50. Perplexity=5 Perplexity=30 Perplexity=100 Dataset: UCI ML digit image data
- 7. Number of iterations (n_iter) required for convergence of the approach Dataset: UCI ML digit image data Parameters of t-SNE (Number of iterations)
- 8. Number of iterations (n_iter) required for convergence of the approach Dataset: UCI ML digit image data n_iter=250 Parameters of t-SNE (Number of iterations)
- 9. 9 Number of iterations (n_iter) required for convergence of the approach Dataset: UCI ML digit image data n_iter=250 n_iter=500 Parameters of t-SNE (Number of iterations)
- 10. Parameters of t-SNE (Number of iterations) 10 Number of iterations (n_iter) required for convergence of the approach Dataset: UCI ML digit image data n_iter=250 n_iter=500 n_iter=2000
- 11. Please share what you learned with others