It'd be nice to support NMF as a model_type, then render only points, where each point represents a document (or subregion of a document if a user specified a window_length argument or similar). By hovering on a point we could fetch the text from that passage:
More generally it'd be nice to create a uniform vector space in which, say, a text region describing cats would hash to the same position as an image of cats. Then the mouseover could allow one to explore a vector space with heterogenous object types...
It'd be nice to support NMF as a model_type, then render only points, where each point represents a document (or subregion of a document if a user specified a window_length argument or similar). By hovering on a point we could fetch the text from that passage:
More generally it'd be nice to create a uniform vector space in which, say, a text region describing cats would hash to the same position as an image of cats. Then the mouseover could allow one to explore a vector space with heterogenous object types...