t-SNE is a tool to visualize high-dimensional data. It converts similarities between data points to joint probabilities and tries to minimize the Kullback-Leibler divergence between the joint probabilities of the low-dimensional embedding and the high-dimensional data.
This project provides some utility functions to apply the t-SNE algorithm implemented in scikit-learn to some PyTorch built-in dataset, e.g., MNIST and FashionMNIST. So you have to install the sklearn and PyTorch first.
python t_SNE.py By default, the FashionMNIST is used and the visualization type is set to image_scatter. You can play with MNIST and use the point_scatter visualization style instead, which correspond to #L221 and #L230, respectively.
- MNIST
- FashionMNIST
Currently, the MNIST and FashionMNIST datasets are supported. And I just simply flatten the spatial dimensions out (i.e., H and W) for an image array. For compact features extracting, one can consider to use neural networks to get semantic features first, and use that to feed t-SNE.