Updated README (2)
An embedding projector using
matplotlib. Allows you to browse embeddings in a 3D scatterplot. You can zoom in and out, dump values to a
pandas-compatible format, &c.
python src/main.py --helpto learn about available options.
python src/main.py $text1 $text2 $text3…to start.
zcoordinates for the embedding.
The above image provides a sense of what one should expect to see with this tool. In it, two books have had their sentences plotted out with GPT-2: one work of fiction and one work of non-fiction. While most of the sentences converge on each other, a clear wedge of sentence extends southwards. Examining this group reveals sentences like chapter labels and page numbers, structural strings whose semantic quality is different than the average sentence in a book. Perhaps not surprising, but certainly nice to see visually.