Buradasın
t-SNE Algorithm Overview
enjoyalgorithms.com/blog/tsne-algorithm-in-ml/Yapay zekadan makale özeti
- Definition and Purpose
- t-SNE is an unsupervised non-linear dimensionality reduction technique
- It preserves similarity between data points in lower dimensions
- Reduces visualization challenges of datasets with more than three features
- Working Process
- Initializes points randomly on a 1D line
- Calculates similarity between points using t-distribution
- Normalizes perplexity scores to make overall similarity sum equal to one
- Creates similarity matrix showing points' cluster membership
- Moves points to align with 2D similarity matrix
- Advantages and Disadvantages
- Handles non-linear data unlike PCA
- Preserves local structure of data points
- Requires complex computational calculations
- Results may vary due to randomization
- Implementation
- Uses Python libraries like pandas and matplotlib
- Applied to MNIST dataset containing 60,000 handwritten digits
- Data needs normalization before training
- Results show similar points grouped together in 2D space