Is it reasonable to use DiffusionMap with n_pcs = NA?

[sam-israel]

The help page on DiffusionMap says that using e.g. 50 PCs results in more regular looking diffusion maps. First, I would like to ask - it results in more regular looking diffusion maps compared to what ?

Second, Is it reasonable not to use projections for the diffusion map (n_pcs=NA), in case that the result looks more reasonable biologically ?