Following on from my previous post on building The Burrito Bot, I want to delve into visualisation of vector embeddings that were generated from the restaurant data pulled from Google Maps.
Those embeddings had 1536 dimensions, each dimension corresponding to an axis within a high dimensional space, with embeddings that have similar meanings grouped together in that high dimensional space.
1536 dimensions…is a lot of dimensions! And for me, a hard concept to get my head around. It all just feels so abstract (to me anyway), I want to see what they actually look like!
Click through for a link to a website that helps with that visualization. It ultimately performs principal component analysis (PCA) to get 1536 (or however many) dimensions down to 3 principal components. It’s not perfect, but it does give us the ability to reason over the data.