These scripts can be used to view what the hidden nodes of a network trained on text data have learned. The scripts take a caffe model as parameters
python tsne_viz.py structure.prototxt model.caffemodel corpus.lex
python tsne_viz.py structure.prototxt model.caffemodel corpus.lex [dopca [perplexity]]
They load the trained caffemodel and give the user options to visualize various parts of the network.